A Thesis submitted in partial fullfilment of the requirements for the degree of

Master of Science in Electronic Engineering

Department of Electronic Engineering
Faculty of Engineering and Applied Sciences
Muhammad Ali Jinnah University, Islamabad

Qadeer Ahmed
June 2009

In the Name of ALLAH, the Most Gracious,

the Most Merciful

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

Helicopter Modeling & Analysis ........................................................ - 6 -

2.1

Helicopter Dynamics ...................................................................................... - 7 -

2.1.1

Elevation Dynamics: ............................................................................... - 8 -

2.1.1.1

Gravitational and Centrifugal Torque ................................................. - 8 -

2.1.1.2

Main rotor Torque ............................................................................... - 8 -

2.1.1.3

Gyroscopic Torque............................................................................ - 10 -

2.1.1.4

Frictional Torque .............................................................................. - 11 -

2.1.2

Azimuth Dynamics ............................................................................... - 12 -

2.1.3

Motor and Rotor Dynamics .................................................................. - 12 -

2.1.4

Sensor Dynamics .................................................................................. - 13 -

2.2

Mathematical model...................................................................................... - 13 -

2.2.1

Non-Linear Model ................................................................................ - 14 -

2.2.2

Linear Model ......................................................................................... - 15 -

2.2.3

Re-Formulation of Mathematical Model .............................................. - 17 -

2.2.4

CE150 Helicopter Model ...................................................................... - 18 -

2.2.5

Physical system description .................................................................. - 19 -

2.2.6

Model Validation .................................................................................. - 21 -

2.3

Model Analysis ............................................................................................. - 21 -

2.3.1

Root locus ............................................................................................. - 22 -

2.3.2

Bode plots ............................................................................................. - 23 -

2.3.3

Singular Values plot .............................................................................. - 24 -

viii

2.3.4

Phase portrait ........................................................................................ - 25 -

2.3.5

Controllability ....................................................................................... - 26 -

2.3.6

Observability ......................................................................................... - 26 -

2.4

Control Challenges........................................................................................ - 27 -

Chapter 3
3.1

Control Algorithms & Decoupling Techniques ................................ - 28 Existing Controllers ...................................................................................... - 29 -

3.1.1

Classical Controllers ............................................................................. - 29 -

3.1.1.1

PID Controller................................................................................... - 29 -

3.1.1.2

State feedback control ....................................................................... - 30 -

3.1.2

Non-Linear Predictive Control ............................................................. - 30 -

3.1.3

Feedback linearization .......................................................................... - 31 -

3.1.4

Time Optimal and Robust control......................................................... - 31 -

3.1.5

Sliding Mode Control ........................................................................... - 32 -

3.1.6

Higher Order sliding mode (HOSM) control ........................................ - 33 -

3.2

Decoupling Techniques ................................................................................ - 34 -

3.2.1

Multi Variable Decouple control .......................................................... - 34 -

3.2.1.1

Boksenbom & Hood Decoupling Technique .................................... - 35 -

3.2.1.2

Zalkin & Lyben Decoupling Technique ........................................... - 36 -

3.2.2

State Space Approach for Decoupling .................................................. - 38 -

3.2.2.1

Static Decoupling .............................................................................. - 38 -

3.2.2.2

Dynamic Decoupling ........................................................................ - 39 -

3.2.3

Near Decoupling Techniques ................................................................ - 40 -

3.2.3.1
3.2.4
3.3

Near-Decoupling: State Feedback .................................................... - 41 Hadamard Weights in LSDP for Robust Decoupling ........................... - 42 -

Conclusion .................................................................................................... - 43 H Controller Design ....................................................................... - 44 -

Chapter 4
4.1

General Control Problem Formulation for H Control ................................. - 45 -

4.2

Mixed sensitivity procedure .......................................................................... - 46 -

4.2.1

Choice of Weights and controller design .............................................. - 48 -

4.2.2

Simulation Results ................................................................................ - 50 -

4.3

Loop shaping design procedure (LSDP) ....................................................... - 51 -

ix

4.3.1

Choice of Weights and Controller Design ............................................ - 52 -

4.3.2

Hadamard weight .................................................................................. - 54 -

4.3.3

Simulations Results............................................................................... - 55 -

4.4

Experimental test Results .............................................................................. - 57 -

4.5

Performance Evaluation ................................................................................ - 62 -

4.6

Conclusion .................................................................................................... - 64 -

Chapter 5

Nonlinear Control Algorithms ......................................................... - 65 -

5.1

Lyapunov Theory .......................................................................................... - 66 -

5.2

Sliding mode control ..................................................................................... - 68 -

5.2.1

Sliding Surface Design 1 ...................................................................... - 69 -

5.2.2

Sliding Surface Design 2 ...................................................................... - 70 -

5.2.2.1

Simulation Results ............................................................................ - 73 -

5.2.2.2

Experimental Test Results ................................................................ - 76 -

5.2.2.3

Performance evaluation .................................................................... - 78 -

5.2.3

Sliding Surface Design via LMIs .......................................................... - 79 -

5.2.3.1
5.3

Simulation Results ............................................................................ - 83 -

Higher order sliding mode control ................................................................ - 84 -

5.3.1

Super Twisting Algorithm .................................................................... - 85 -

5.3.1.1

Simulation Results ............................................................................ - 88 -

5.3.1.2

Experimental Results ........................................................................ - 89 -

Chapter 6

Conclusion & Future Work............................................................. - 94 -

Chapter 7

Appendix

................................................... - 98 -

7.1

MATLAB code of modeling of Helicopter Model ....................................... - 99 -

7.2

MATLAB code for H with Traditional weights ....................................... - 100 -

7.3

MATLAB code for H with Hadamard weights ........................................ - 102 -

7.4

SIMULINK Diagram for H Implementation on Helicopter Model .......... - 104 -

7.5

SIMULINK Diagram for H Implementation on Helicopter Model .......... - 105 -

7.6

SIMULINK Block Diagram for 2-SMC ..................................................... - 108 -

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.

2.1 Helicopter Dynamics

The helicopter dynamics can be reduced to vertical and horizontal plane dynamics
which can be approximated as twin rotor dynamics as shown in Figure 2-1. An attempt to
model the system dynamics in detail leads to extremely complicated, not readable and not
useful model. In our case the model will be used for investigating the system dynamics
with respect to control tasks. The system will operate in some working conditions only
and not all of the dynamical properties will be invoked. This leads to the assumptions
which will simplify the derivation of the model. We propose two ways model the system.
The first one is a systematic modeling method based on variational approach, i.e.
Lagrange's equations. The second approach is a direct derivation of the model by
computing the force balances. Both methods lead to the model in the form of nonlinear
differential equations [7]. System identification techniques are also used to model the
system, but the resultant model has linear dynamics only thus prevent us to understand
actual working of the system in detail. Several other authors have discussed the helicopter
dynamics in detail in new approach, the details of which can be found in [8].

Figure 2-1: Symbolic representation of helicopter with 2-DOF

___________________________________- 7 -_______________________________

2.1.1 Elevation Dynamics:

The modeling in elevation of helicopter is carried out using standard physics laws of
angular momentum. Considering the free body diagram of helicopter model, different
torques produced by different forces are balanced about the pivot point. The different
torques produced are
a) Gravitational Torque
b) Frictional Torque
c) Centrifugal Torque
d) Main rotor Torque
e) Gyroscopic Torque

2.1.1.1 Gravitational and Centrifugal Torque

Consider the free body diagram shown in Figure 2-2, the weight of the helicopter
and centrifugal force produce respective torques about the pivot point. Eq. (2.1) describes
the gravitational torque produced by the model weight.

w l w sin

(2.1)

w Gravitational torque ( N .m)

Elevation Angle (rad )

Where

w Weight of helicopter (kg.m / sec 2 )

l Moment Arm (m)

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)

AngularVelocity in horizontal plane (rad / sec)

Fc Centrifugal Force ( N )

2.1.1.2 Main rotor Torque

The main rotor force produced is the consequence of its angular speed as shown
in Figure 2-3. The more the angular speed of rotor the more force will be induced on the

___________________________________- 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)

F1 Main rotor Force ( N )

Where

1 Main rotor angular velocity (rad / sec)

1 Main rotor torque ( N .m)

We can calculate the force caused by the angular velocity as,

F1 mr a
v2
r
v r
a

a r 2

F1 mr r 2
Where

(2.5)

r= radius of main rotor (m)

mr = mass of main rotor (kg )

Finally we have the main rotor torque as

1 k1 2
Where

(2.6)

k1 m r l

___________________________________- 9 -_______________________________

Figure 2-3: Main rotor torque caused by 1 in vertical plane

2.1.1.3 Gyroscopic Torque

Gyroscopic torque occurs as a result of Coriolis forces acting on helicopter
elevation dynamics. This torque results when moving main rotor changes its position in
azimuth. Thus resultant gyroscopic torque caused by the main rotor and azimuth rotation
can be calculated from the Figure 2-4 as
cos
G = k2
1

(2.7)

= Azimuth Angle (rad )

Where

= Angular velocity in Azimuth (rad / sec)

k2 = Contant of proportionality ( N .m / sec)

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 .

2.1.1.4 Frictional Torque

The frictional torque can be estimated from the following equation

f = B1

(2.8)

B1 = Damping Constant (kg .m 2 / sec)

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)

I1 moment of inertia of the helicopter body around horizontal axis (kg.m 2 )

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 -_______________________________

2.1.2 Azimuth Dynamics

The net torques produced in horizontal plane as seen in Figure 2-6 are
I 2 2 r f

(2.10)

2 Side rotor torque ( N .m)

f Frictional troque ( N .m)
2

Where

r Main motor reaction torque ( N .m)

I 2 Moment of Inertia in verticle plane ( Kg.m 2 )

Figure 2-6: Mechanical Torques produced in horizontal plane

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)

2.1.3 Motor and Rotor Dynamics

The details of motors modeling are available in [7]., in our case DC motor has
been estimated against a simple first order transfer function whose time constant has been
identified by motor behavior. Eq. (2.12) shows the transfer function of motor

___________________________________- 12 -_______________________________

M1

1
T1s 1

(2.12)

T1 Main motor time constant

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)

u1 Output of the motor

Where

2.1.4 Sensor Dynamics

Incremental sensor is installed in the model to measure the angular position in
elevation and azimuth. These parts have no dynamics and are considered to be linear in
the whole extent of measured angles. The 10-bit encoder gives 1024 pulse against 1
degree and the encoder can be modeled with the following equations
y k yo

(2.14)

y k

(2.15)

y Elevation output angle

y Initial Elevation output angle
o

Where

y Azimuth output angle

k Elevation constant
k Azimuth constant

2.2 Mathematical model

The above discussed dynamics of different components of helicopter model help us
in formulating the mathematical model which can be seen in Figure 2-7. The different
states of the model can be figured out on the basis of their function, like each motor has
one state that will give the information of angular speed of the motor, similarly the
angular position and rate of change of angular positions give rise to few other states.
Finally our systems states and outputs come out to be

___________________________________- 13 -_______________________________

Main motor speed

x1 1

Elevation Angle

x2

x3
Angular speed in Elevation

Side motor speed
x4 2

x
Azimuth
Angle

Angular speed in Azimuth

x6

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.

Figure 2-7: Block diagram of nonlinear model of twin rotor system

2.2.1 Non-Linear Model

The non-linear model based on above states can be developed from the basic
equation derived from physical laws is,
x1 =

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]..

2.2.2 Linear Model

The non-linear model can be linearized around a set point to develop linear model
that would be functional in linear range. Conventional linearization technique by taking
the Jacobean of non-linear model and replacing the set points values has been utilized to
develop the linear model. Eq. (2.23) explains the linearization procedure mathematically

f ( x)
x x 0

(2.23)

For example we can linearize Eq. (2.19) as

x3 =

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

Finally the linearized model of helicopter is formulated as below

___________________________________- 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 )

7.141 s 4 + 53.07 s3 + 118 s 2 + 79.94 s

s 7 + 11.19 s 6 + 54.6 s 5 + 178 s 4 + 427.2 s 3 + 596.6 s 2 + 327 s

g12 ( s )

-1.776e-15 s 6 + 5.684e-14 s3 + 1.137e-13 s 2

s 7 + 11.19 s 6 + 54.6 s5 + 178 s 4 + 427.2 s3 + 596.6 s 2 + 327 s

g 21 ( s )

3.553e-15 s 6 - 1.401 s5 - 11.37 s 4 - 39.24 s3 - 110.8 s 2 - 199.4 s - 59.68

s 7 + 11.19 s 6 + 54.6 s5 + 178 s 4 + 427.2 s3 + 596.6 s 2 + 327 s

g 22 ( s )

-1.776e-15 s 6 + 28.41 s 4 + 144.5 s3 + 431 s 2 + 1215 s + 1106

s 7 + 11.19 s 6 + 54.6 s5 + 178 s 4 + 427.2 s3 + 596.6 s 2 + 327 s

2.2.3 Re-Formulation of Mathematical Model

The differential model obtained in Eq.(2.18) Eq. (2.22) can be reformulated in
standard mechanical systems dynamical equation as

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

The matrices values for helicopter model are

I
M 1
0

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).

2.2.4 CE150 Helicopter Model

The CE150 Helicopter Model shown in Figure 2-8 is designed for the theoretical
study and practical investigation of basic and advanced control engineering principles.
This includes system dynamics modeling, identification, analysis and various controllers
design by classical and modern methods [7]. The twin rotor system resembles a
simplified behavior of a real helicopter with fewer degrees of freedom. In real helicopters
the control is generally achieved by tilting appropriately the blades of the rotors with the
collective and cyclic actuators, while keeping constant rotor speed. In order to simplify
the mechanical design of the system, the laboratory setup employed, is designed slightly
differently. In this case, the blades of the rotors have a fixed angle of attack, and control
is achieved by controlling the speeds of the rotors. As a first consequence of this, the
laboratory helicopter presents higher coupling between dynamics of the rigid body and
dynamics of the rotors than a conventional helicopter, 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]. In order to
elaborate the coupling affect in dynamics, while implementation this thesis involves the
practice of first restricting twin rotor system to 1-DOF (Elevation) and after some time 2DOF (Azimuth) is introduced to exaggerate coupling effects.

___________________________________- 18 -_______________________________

Figure 2-8: CE150 HUMUSOFT Helicopter Model

Figure 2-9: Schematic diagram of helicopter model

2.2.5 Physical system description

The laboratory helicopter is commonly known as twin rotor multi input multi
output (MIMO) system (TRMS). This system is hinged as the based, thus restricting the
six degrees of motion to just two degrees of freedom. This educational model consists of
two DC motors which drive upper and side propeller by generating torques perpendicular
to their rotation and a servo mechanism used to manipulate the center of gravity. The
system has two degrees of freedom i.e. Elevation ( ) in vertical plane and azimuth ( ) in
horizontal plane, which are measured precisely by incremental encoders installed inside

___________________________________- 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

DC motor with permanent magnet

Max Voltage 12V
Max Speed 9000 RPM

Side Motor 2

DC motor with permanent magnet

Max Voltage 6V
Max Speed 12000 RPM

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

Table 2-1 System specifications (HUMUSOFT CE 150 Manual)

2.2.6 Model Validation

The linear model extracted from the non-linear model is validated against impulse
signal. The elevation dynamics in vertical plane are asymptotically stable in open loop
and can be verified in open loop against any validation signal. However, azimuth
dynamics in horizontal plane are not asymptotically stable and can be verified by
trapping it in feedback to make it asymptotically stable and applying any validating
signal with proportional gain in the forward path. Figure 2-10 shows the validation of
linear model against impulse signal given for 2 seconds at input of the elevation in open
loop. The validation response clearly shows the non-linearities impact which were
ignored during linearization e.g. the gravitational torque caused by weight of the body is
non-linear terms, its difference is clearly visible in impulse response when helicopter
declines after the impulse input ends.
Impulse response
0.12
Actual Response
Linear model Response
Error

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

2.3 Model Analysis

The linear model and non-linear model are ready for analysis. Table 2-2 shows the
Eigenvalues of the linear model and their properties

___________________________________- 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

Table 2-2: Eigenvalues of helicopter model

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.

2.3.1 Root locus

The root locus [18] of each component of G(s) is shown in Figure 2-11. The
elevation root locus gives the picture that small increase in gain will make the system to
go unstable by shifting the poles on right half plane. The second and fourth root locus
point out the right half plane zeros, which make s our system non-minimum phase
system. After going through root locus we clearly get the idea that we cannot give high
gain to our system and we have to be careful about right plane zeros in future designing
procedures.

___________________________________- 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

Figure 2-11: Root locus of each SISO systems given in G(s)

2.3.2 Bode plots

The bode plots [18] shown in Figure 2-12 give us the idea of each component of
G(s). The first bode show the behavior of 1st input to 1st output, second bode plot show us
the 1st output and 2nd input transfer function behavior and so on. The helicopter dynamics
in elevation have poor gain and phase margins however these margins are fine in azimuth
dynamics. The effect of 1st input on azimuth is significant at lower frequencies but the
elevation has very less impact of 2nd input. The bode plots give us the idea of coupling in
helicopter dynamics which will be discussed further in detail.

___________________________________- 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)

Figure 2-12: Bode plots of SISO systems given in G(s)

2.3.3 Singular Values plot

The singular values [19] shown in Figure 2-13 exhibits the poor tracking of the
system and the slope of the singular values at zero crossing is around 2 which founds the
base for poor tracking performance. The slight peak at zero crossing is also indicating the
oscillatory behavior of the helicopter. However, the upper and lower singular values at
higher frequencies are desirable.

___________________________________- 24 -_______________________________

Singular Values
60

40

20

Singular Values (dB)

-20

-40

-60

-80

-100

-120
-2
10

10

-1

10

10

10

Frequency (rad/sec)

Figure 2-13: Singular values of Twin Rotor MIMO System

2.3.4 Phase portrait

Phase portrait [6] give us the clear picture that helicopter dynamics in vertical
plane are inherently asymptotically stable and elevation eventually reaches to zero with in
finite time whatever the initial conditions are given. The elevation phase portrait can be
seen in Figure 2-14.
Phase portrait of Elevation
0.4

0.3

Rate of change of elevation

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

Figure 2-14: Phase portrait of Elevation

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

Rate of Change of Azimuth

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

Figure 2-15: Phase portrait of 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.

2.4 Control Challenges

The major challenges determined from the above analysis are tracking control and
reduction of coupling with in the system dynamics. The tracking problem can be seen
from singular values shown in Figure 2-13, the lower singular values have poor gains at
lower frequencies and the upper and lower singular values have slope around 2 at zero
crossing, these two characteristics of model ensures the poor tracking response of
helicopter model. The stability constraints are visible in root locus i.e. we cannot have
high gains for controllers. Similarly the non-minimum phase behavior is also to be kept
in mind while designing and testing controllers of CE 150 helicopter model.
The major challenge residing in the model is interstate coupling. Eq. (2.19) in nonlinear model shows the effect of horizontal plane dynamics in elevation caused by
gyroscopic torque. This coupling is not as significant as the coupling is induced from 1st
input to azimuth. Eq. (2.22) in non-linear model shows the clear effect of it. In fact this
coupling has been indentified in detail in the helicopter manual [7]. This coupling affects
the performance of model especially in horizontal plane dynamics.
After discussing the helicopter model in detail and figuring out its control objectives,
now the control and decoupling techniques will be explored. Several attempts have been
made to achieve the control objectives which will be discussed in the coming chapters.
Secondly, to over come the coupling problem various decoupling techniques are
discussed to handle the interstate coupling which affects the helicopter model
performance.

___________________________________- 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.

3.1 Existing Controllers

Controllers ranging from linear to nonlinear domain have been validated on this
model. The educational manual accompanied with the model has some classical control
design techniques. Moreover, several other control engineers have validated their
designed control algorithms on this model. This section will discuss an overview of some
of the previously designed controllers for this model.

3.1.1 Classical Controllers

The classical controllers, the control techniques invented in early 1960s are
termed as classical control techniques. Few of these techniques have been implemented
on this model which includes PID and state feedback controllers. Although these
techniques give good performance but under uncertain conditions these techniques fail to
cope up with desired performance.

3.1.1.1 PID Controller

The gains of proportional, integral and derivative constant in PID controller are
extracted from the root locus, shown in Figure 2-11. The root locus from main motor as
actuator and elevation as measured output will give the limits of PID gains which will
ensure the systems eigenvalues in the left half plane similarly the PID gains for side
motor as actuator and azimuth as output can be obtained from its root locus. More details
___________________________________- 29 -_______________________________

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.

3.1.1.2 State feedback control

The second classical control technique which has been applied for the helicopter
tracking control is pole placement using state feedback control. The main idea of this
method is that the location of the closed loop poles of a linear system determines its
dynamics. In this problem the poles of a closed loop system are placed according to the
position specified by the designer. This may be done only by the state feedback. Eq.(3.2)
delivers the mathematical concept of pole placement using state feedback, where K is the
feedback controller which places the closed loop system poles at desired location to get
the preferred performance.
X AX BU
U KX
X ( A BK ) X

(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].

3.1.2 Non-Linear Predictive Control

The model predictive control [20] problem is formulated as solving on-line a
finite horizon open-loop optimal control problem subject to system dynamics and
constraints involving states and controls. Based on the measurements obtained at certain
time, the controller predicts the future dynamic behavior of the system over a prediction
horizon and determines the input such that a predetermined open-loop performance

___________________________________- 30 -_______________________________

objective functional is optimized. Dutka et al [21] have implemented nonlinear predictive

control for tracking control of helicopter model. The nonlinear algorithm is based on
state-space generalized predictive control. The non-linearity is handled by converting the
state-dependent state-space representation into the linear time-varying representation.
The predictions of the future controls are used to calculate predictions of the future states
and future time varying system parameters. Applied to the helicopter model, the
algorithm performs well. It is capable of the stabilizing the system for maneuvers for
which its linear counterpart fails. The authors have only addressed the tracking
performance which is very slow as its settling time is high and secondly the coupling
effect on its performance is not considered.

3.1.3 Feedback linearization

Feedback linearization [6] offers the user to cancel out the nonlinearities and
introduce Hurwitz polynomial that will be responsible for control of system. Based on
system properties one has to decide whether to use exact feedback linearization, input
output linearization or input state linearization. M.Lopez et al [9], [10], [11] have
presented control of helicopter model using feedback linearization. The feedback
linearization techniques are carried out in elevation dynamics and azimuth dynamics are
kept at zero. The approximate feedback linearization of elevation dynamics is performed
by means of over parameterization of elevation model. After the linearization, LQR
robust control algorithm has been implemented to deliver desired results. In the second
attempt, the authors have used input output linearization technique to linearize elevation
dynamics; it was presented as the input state linearization was not suitable when the
velocity of the rotors were next to zero, since the control law saturated. The fore
discussed techniques have presented tracking control of the helicopter elevation dynamics
and coupling effect was ignored as the azimuth dynamics were kept at zero.

3.1.4 Time Optimal and Robust control

Te-Wei et al [12] proposed time optimal control for helicopter model. The MIMO
system is first decomposed in two SISO systems and coupling effect is taken as
disturbances or change of system parameters. For each SISO system optimal control has
been designed which can tolerate 50% changes in the system parameters; however the
___________________________________- 31 -_______________________________

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.

3.1.5 Sliding Mode Control

In sliding-mode control [6] design a hyper-plane is defined as a sliding-surface.
This design approach comprises of two stages; first is the reaching phase and second is
the sliding phase. In the reaching phase, states are driven to a stable manifold by the help
of appropriate equivalent control law and in the sliding phase states slide to an
equilibrium point. One advantage of this design approach is that the effect of nonlinear
terms which may be construed as a disturbance or uncertainty in the nominal plant has
been completely rejected. Another benefit accruing from this situation is that the system
is forced to behave as a reduced order system; this guarantees absence of overshoot while
attempting to regulate the system from an arbitrary initial condition to the designed
equilibrium point. Koudela et al [22] designed sliding mode control for the regulation

___________________________________- 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

3.1.6 Higher Order sliding mode (HOSM) control

Higher order sliding mode control [24], [25], [26]is a recent approach that allows
the removal of all standard sliding mode restrictions, while preserving the main sliding
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

___________________________________- 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.

3.2 Decoupling Techniques

Decoupling techniques involves a procedure which helps us to overcome couplings
residing in the system dynamics. This cross-couplings effect desired performance of the
system. Decoupling in control theory can be achieved implicitly by declaring coupling as
disturbance and handle it with robust controller or explicitly by introducing decouplers
which negate the couplings affects.
The coupling in the helicopter model has discussed in detail in Section 2.3 and
Section 2.4. The control techniques that have been implemented on the model have the
objective to extract the tracking performance from the helicopter model without handling
the coupling effects on the system dynamics. Therefore in order to get maximum tracking
performance and to overcome the undesirable coupling dynamics several decoupling
techniques have been explored. All of the following discussed techniques have been
attempted to decouple our system but few of the techniques could deliver desired results.
Other algorithms were unsuccessful to deliver results at certain stage.

3.2.1 Multi Variable Decouple control

Another popular approach for dealing with control loop interactions is to design
on interacting or decoupling control schemes. Here, the objective is to eliminate
completely the effects of loop interactions. This is achieved via the specification of
compensation networks known as decouplers. Essentially, the role of decouplers is to
decompose a multivariable process into a series of independent single-loop sub-systems.
If such a situation can be achieved, then complete or ideal decoupling occurs and the
multivariable process can be controlled using independent loop controllers [28]. The
Figure 3-1 shows the general decoupling control structure.

___________________________________- 34 -_______________________________

Figure 3-1: General Structure to decouple the system

The idea discussed can be implemented in the following two techniques

3.2.1.1 Boksenbom & Hood Decoupling Technique

The Boksenbom and Hood technique [28] takes the decoupler as shown in Figure
3-2. The system will be decoupled if the off diagonal terms are forced to zero and only
diagonal terms are there in the pseudo plant.

Figure 3-2: Decoupling control system (Boksenbom and Hood)

This can be achieved as

GGc* diag (q1 , q2 )

(3.4)

The matrix product of the two multivariable transfer functions will as

___________________________________- 35 -_______________________________

G12 Gc*,11 Gc*,12 q1

G
GGc* 11

*
*
G21 G22 Gc ,21 Gc ,22 0

0
q2

G11Gc*,11 G12Gc*,21 G11Gc*,12 G12Gc*,22 q1

GG

*
*
*
*
G21Gc ,11 G22Gc ,21 G21Gc ,12 G22Gc ,22 0
*
c

(3.5)

0
q2

(3.6)

From Eq. (3.6) we can say

q1 G11Gc*,11 G12Gc*,21

(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.

3.2.1.2 Zalkin & Lyben Decoupling Technique

The second scheme for decoupling multi variable systems is shown in Figure 3-3,
which is known as Zalkin and Lyben decoupling technique [28]. Here, in addition to the
decoupling network, there are two extra blocks which represent the forward path
controllers. In contrast to the previous strategy, the decoupling network forms the
secondary post-compensation block and allows more flexibility in the implementation
and commissioning of the non-interacting control scheme.
The decoupler values in the forward can be found by using linear algebra
techniques as

___________________________________- 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)

Since X diag ( x1 , x2 ) , then

Gc* G 1 X

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)

This leads to the off diagonal terms as

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 -_______________________________

Figure 3-3: Non-interacting decoupling control structure (Zalkind &Luyben)

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.

3.2.2 State Space Approach for Decoupling

Consider the linear system

x Ax Bu
y Cx Du

(3.20)

Where x R n , u, y R m , A R n n , B R n m , C R m n , D R mm . The transfer function of

system in Eq. (3.20) can be written as

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

3.2.2.1 Static Decoupling

A system is said to be statically decoupled if it is stable and its static gain matrix
G(0) is diagonal and non-singular. The control law for system in Eq. (3.20) can be taken
as

___________________________________- 38 -_______________________________

u Kx Fr

(3.22)

Such that the close loop transfer function

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.

Design K which ensures the roots of ( sI A BK ) in stable region.

ii.

Obtain F as F [(C DK )( sI A BK ) 1 B D]1

iii.

Design control law as given in Eq.(3.22).

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].

3.2.2.2 Dynamic Decoupling

The system in Eq.(3.20) is said to be dynamically decoupled by the control law in
Eq. (3.22) if and only if the matrix B* exist and it is non-singular. If this is the case, by
choosing

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].

3.2.3 Near Decoupling Techniques

Several approaches for decoupling control have been discussed in the previous
sections. Robust stability and plant uncertainties are not addressed explicitly in design. It
is noted that exact decoupling may not be necessary even in the nominal case for many
applications and usually impossible in the uncertain case. What is really needed instead is
to limit the loop interaction to an acceptable level. In this section the concept of neardecoupling and development, an approach to near-decoupling controller design for both
nominal and uncertain systems is discussed. Two techniques are discussed here in brief.
Details and rest of the techniques can be found in [29].
Consider the system defined in Eq. (3.20) where the B,C,D can be split into p partitions.

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

i {1, 2..... p}, [, ]

i, j {1, 2,..... p}, i j

holds for given numbers of i 0 , i 1, 2.... p and 0

3.2.3.1 Near-Decoupling: State Feedback

Consider the system defined by Eq.(3.29), this system is said to be near statefeedback decouplable if there exists matrices Q 0 and F such that the following LMIs
hold
QAT AQ F T BT BF
T
T
T
T
T T
Bi Dii (QCi F Di )

QAT AQ F T BT BF

BTj

(QCiT F T DiT )T

Bi (QCiT F T DiT ) Dii

0
i2 I DiiT Dii

i {1, 2.....m}

Bj
I
Dij

QCiT F T DiT

DijT
0

(3.31)

(3.32)

i, j {1, 2.....m}, i j

The state feedback law is given by

u Kx

(3.33)

K FQ 1

(3.34)

And the close loop performance indices satisfy

___________________________________- 41 -_______________________________

 ( H cl ,ii ( j )) i
H cl ,ij

i {1, 2..... p}, [, ]

(3.35)

i, j {1, 2,..... p}, i j

(3.36)

The robust near decoupling and output feedback decoupling algorithms can be found with
detail in [29].

3.2.4 Hadamard Weights in LSDP for Robust Decoupling

Loop shaping procedure involves the reshaping of system dynamics by
introducing additional transfer functions i.e. weights in the loop. Loop shaping design
procedure (LSDP) allows the user to modify the system by increasing the system gains at
lower frequencies, decreasing the gain at higher frequencies and making the slope around
1 at cross-over frequency, thus making the system to work according to user desires.
These weights later on help us in formulating H controller and these weights are merged
with controller in the implementation. The weights are dependent on the system inputs
and outputs for example if we have two inputs and two outputs system the weight
dimensions will be [2 2] . Normally, these weights are introduced to the system through
matrix multiplication which relates each component of the weights to each component of
plant dynamics as
W W12
W 11

W21 W22

(3.37)

G12 W11 W12 G11W11 G12W21 G11W12 G12W22

G
G W 11

G21 G22 W21 W22 G21W11 G22W21 G21W11 G22W21

(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

G21 G22 W21 W22 G21W21 G22W22

(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 -_______________________________

A control system is robust if it remains stable and achieves certain performance

criteria in the presence of possible uncertainties. The robust design is to find a controller,
for a given system, such that the closed-loop system is robust. The H optimization
approach, being developed in the last two decades has been shown to be an effective and
efficient robust design method for linear, time invariant control systems. Various robust
stability considerations and nominal performance requirements were formulated as a
minimization problem of the infinitive norm of a closed-loop transfer function matrix.
Hence H optimization approach solves, in general, the robust stabilization problems and
nominal performance designs [17]. In this chapter, we shall discuss formulation of a
robust design problem for our helicopter model.

4.1 General Control Problem Formulation for H Control

There are many ways in which feedback design problems can be cast as H
optimization problems. It is very useful therefore to have a standard problem formulation
[19] shown in Figure 4-1 into which any particular problem may be manipulated.

Figure 4-1 General control configuration for H control

The system described in Figure 4-1 can be written as

z
w P11 ( s ) P12 ( s ) w
v P( s) u P ( s) P ( s) u

21

22

(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

Fl ( P, K ) P11 P12 K ( I P22 K ) 1 P21

The H control involves the minimization of infinity norm of Fl ( P, K ) which means to
find the stabilizing controllers K which minimizes
Fl ( P, K )

max ( Fl ( P, K )( j ))

(4.4)

This can interpreted as the performance parameter which involves minimization of

maximum singular value of Fl ( P, K ) [19].Practically, the sub optimal controller for H
problem is simpler to design therefore, if we assume min the minimum value of
Fl ( P, K ) over all stabilizing controllers K, then the sub-optimal control problem is

given by a min and the sub-optimal problem can be formulated as

Fl ( P, K )

(4.5)

This can be efficiently solved using the algorithm discussed in [4].

4.2 Mixed sensitivity procedure

Mixed sensitivity [19] is the name given to transfer function shaping problems in
which the sensitivity function S I GK is shaped along with one or more other
1

closed loop transfer functions such as KS

or the complementary sensitivity

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.

Figure 4-2: One degree of freedom configuration

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.

4.2.1 Choice of Weights and controller design

The weighting function W1 is used to impose performance specifications to the
system. According to Eq. (4.6) it is necessary that So should be having smaller gains at
___________________________________- 48 -_______________________________

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

Singular Values (dB)

-50

-60

-70

-80

-90

-100
-3
10

10

-2

10

-1

10

10

10

10

Frequency (rad/sec)

Figure 4-4 W1, Weighting function for S

The objective of W2 is to reduce overshoots of the temporal response without changing

too much rise time. W2 is typically scalar high pass filter with the cross over frequency
approximately equal to required bandwidth. Figure 4-5 shows the selected W2 for mixed
sensitivity H solution.
Singular Values
-60

-80

Singular Values (dB)

-100

-120

-140

-160

-180
-2
10

10

-1

10

10

10

10

10

10

10

10

10

Frequency (rad/sec)

Figure 4-5 W2, Weighting function for KS

___________________________________- 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.

4.2.2 Simulation Results

The suboptimal controller applied for tracking control on helicopter model
delivered the step responses shown in Figure 4-6. Although, the tracking has been
achieved in 2 seconds but still the coupling effect can be seen in the start for half a
second which serves as barrier to achieve the desired controlled response. The respective
control effort is shown in Figure 4-7.
Step response
From: In(1)

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)

Figure 4-6 Step response (Mixed sensitivity)

___________________________________- 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)

Figure 4-7 Control Signal (Mixed sensitivity)

4.3 Loop shaping design procedure (LSDP)

The second design procedure for H is LSDP. Loop shaping is essentially a two
stage design procedure. First, the open-loop plant is augmented by pre and postcompensator to give a desired shape to the singular values of the open-loop frequency
response. Then the resulting shaped plant is robustly stabilized with respect to co-prime
factor uncertainty using H optimization [19].
For loop shaping design procedure, the stabilization of the plant G which has a
normalized left co-prime factorization [19].
G M 1 N

(4.11)

Figure 4-8: H Robust Stabilization Problem

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)

The maximum value of can be obtained from

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)

will give the optimal/suboptimal controller for loop shaping design.

4.3.1 Choice of Weights and Controller Design

The singular values of the plant are shown in Figure 2-13. To align the singular
values in the frequency range of interest, pre-compensator and post-compensator are
design requirements. Specifications on the closed loop system are indirectly imposed by
selecting the weighting function W1 and W2 and appropriate adjusting the open-loop
frequency response [19]. High gain at low frequencies of the shaped plant implies good
tracking ability. Similarly low gain at higher frequencies implies better noise rejection.
The slope of singular values at crossover frequency also affects the tracking; if the slope
is not 1 then the tracking performance is not guaranteed.

___________________________________- 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].

Figure 4-9 LSDP implementation

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

Singular Values (dB)

50

-50

-100

-150

-200
-3
10

10

-2

10

-1

10

10

10

10

Frequency (rad/sec)

Figure 4-10: Modified Singular Values using Traditional Weights in LSDP

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.

4.3.2 Hadamard weight

Hadamard weights are used in loop shaping design procedure for element by
element weighting in multivariable system which helps in decoupling the plant behavior.
The procedure of integrating weight dynamics to change the system dynamics described
in Eq.(3.39). The weight transfer function used for helicopter model is
1980 s 2 4356 s 792

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

Singular Values (dB)

50

-50

-100

-150
-3
10

10

-2

10

-1

10

10

10

10

Frequency (rad/sec)

Figure 4-11: Modified Singular Values with Hadamard Weights in LSDP

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.

4.3.3 Simulations Results

In the first phase, Matlab based simulations are carried out. The sub optimal
controller with min 2.5914 , designed using weighting procedure as described in Eq.
(3.38) yields acceptable results as shown in Figure 4-12 and control effort required to
obtain these results is shown in Figure 4-13. It can be observed in results that the desired
elevation and azimuth angles are achieved in finite time but in the same time coupling is
also noticeable for few seconds in the start of simulation. So to reduce the coupling we
will have to increase the robustness, the redesigning of the weight will provide more
robustness but the performance will deteriorate. So the designed weight in Eq.(4.18)
delivers the optimum results which reduces coupling to a certain level at the same time
generates the optimal performance 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

Figure 4-13: Control Effort of Traditional Weighted H controller

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)

Figure 4-15: Control Effort of Hadamard Weighted H controller

4.4 Experimental test Results

After the simulations have been carried out, the controller is now implemented at
the actual model. The weights designed in Eq. (4.18) and Eq. (4.19) have been attained
after number of attempts because there were many other weights which delivered ideal
simulation but when it comes implementation the controllers designed on those weights
failed to deliver desired results. The rid tests are carried out to verify various results.
Figure 4-16 shows the step response of helicopter model to acquire equilibrium position,
when implemented with H controller having high robustness i.e. min 2.5914 and

___________________________________- 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 -_______________________________

Control Effort (Normal Weights)

1

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 -_______________________________

Control effort (Hadamard Weights)

Control Effort (Elevation)

1
0.8
0.6

0.4
0.2

10

15

20

10

15

20

25

30

35

40

45

30

35

40

45

Control Effort (Azimuth)

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

Multi Step Response (Hadamard Wt)

25

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

4.5 Performance Evaluation

The analytical performance evaluation of controller based on the output can be
carried out by using performance indices [30]. A performance index is the quantitative
measure of the performance of the system. The four performance indices are defined
based on error e y r dynamics.

___________________________________- 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.

5.1 Lyapunov Theory

Lyapunov theory [6] helps us to develop control algorithms from the equations that
give the idea about the energy of the system. The rate of change of energy, if negative
definite, ensures the system stability. Mathematically, we can say that if energy function
is positive definite V ( x) 0 and rate of change of energy function is negative definite
V ( x) 0 then the system stability is ensured [6].

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)

And for azimuth dynamics the Lyapunov function can be written as

1
V ( x) ( x42 x52 x62 )
2

(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)

Plugging Eq.(2.18) and Eq. (2.19) in Eq. (5.3) we will get

1
1
V ( x) x1 ( ( x1 u1 )) x2 x3 x3 ( ((a1 x1 ) 2 b1 x1 B1 x3 Tg sin x2 K gyro u1 x6 cos x2 )) (5.4)
T1
I1
The control law in vertical plane dynamics comes out to be
u1

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.

5.2 Sliding mode control

In control theory, sliding mode control, or SMC, is a form of variable structure
control (VSC). It is a nonlinear control method that alters the dynamics of a nonlinear
system by application of a high-frequency switching control. The state-feedback control
law is not a continuous function of time; it switches from one smooth condition to
another. That is, the structure of the control law changes based on the position of the state
trajectory; hence, sliding mode control is a variable structure control method because it
switches from one smooth control law to another. The multiple control structures are
designed so that trajectories always move toward a switching condition, and so the
ultimate trajectory will not exist entirely within one control structure. Instead, the
ultimate trajectory will slide along defined manifolds. The motion of the system as it
slides along these boundaries is called a sliding mode [6].
Intuitively, sliding mode control uses practically high gain to force the trajectories
of a dynamic system to slide along the restricted sliding mode subspace. Trajectories
from this reduced-order sliding mode have desirable properties (e.g., the system naturally
slides along it until it comes to rest at a desired equilibrium). The main strength of sliding
mode control is its robustness. Because the control can be as simple as a switching
between two states (e.g., "on"/"off" or "forward"/"reverse"), it need not be precise and
will not be sensitive to parameter variations that enter into the control channel.
Additionally, as the control law is not a continuous function, the sliding mode can be
reached in finite time (i.e., better than asymptotic behavior). In particular, because
actuators have delays and other imperfections, the hard sliding-mode-control action can
lead to chatter, energy loss, plant damage, and excitation of unmodeled dynamics.
The design process of sliding-mode controller involves a hyper-plane defined as a
sliding-surface. This design approach comprises of two stages; first is the reaching phase
and second is the sliding phase. In the reaching phase, states are driven to a stable
manifold by the help of appropriate equivalent control law and in the sliding phase states
slide to an equilibrium point. One advantage of this design approach is that the effect of

___________________________________- 68 -_______________________________

nonlinear terms which may be construed as a disturbance or uncertainty in the nominal

plant has been completely rejected. Another benefit accruing from this situation is that
the system is forced to behave as a reduced order system; this guarantees absence of
overshoot while attempting to regulate the system from an arbitrary initial condition to
the designed equilibrium point.

5.2.1 Sliding Surface Design 1

The sliding manifold based on elevation states in Eq. (2.16) can be chosen as
s1 ( x) x3 x2

(5.11)

The positive definite Lyapunov function for sliding manifold can be chosen as
V ( s)

1 2
s1
2

(5.12)

The time derivative of the Lyapunov function can be written as

V ( s ) s1s1

(5.13)

s1 ( x) x2 x3

(5.14)

where

Plugging the values of Eq.(2.19) in Eq.(5.14), we get

s1 ( x) x3 (

1
((a1 x1 ) 2 b1 x1 B1 x3 Tg sin x2 K gyro u1 x6 cos x2 )
I1

(5.15)

Putting Eq.(5.15) in Eq. (5.13) we get

1
V ( s ) s1 ( x3 ( ((a1 x1 ) 2 b1 x1 B1 x3 Tg sin x2 K gyrou1 x6 cos x2 ))
I1

(5.16)

From Eq.(5.15) we can derive the equivalent control law ueq as

ueq

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)

Eq.(5.13) that ensures the system stability, comes out to be

___________________________________- 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)

5.2.2 Sliding Surface Design 2

In the second attempt to design the sliding mode control for twin rotor system is
carried out by splitting the MIMO system into two SISO systems, for each SISO system
the sliding manifolds are designed based on their error dynamics defined in Eq. (5.25)
E X X eq

(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 -_______________________________

The above system in Eq. (5.26) can be rewritten as

e3 s1 f (e1 , e2 )
e s f (e , e )
4 5
6 2

(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

The chattering reduction depends on value of 1 . The chattering reduction depends

on value of 1 but at the cost of robustness, the more the value of , the lesser will
be chattering but at the same time robustness will be reduced. As we know, angular
positions and velocities in the dynamical model always remain bounded due to the
mechanical structure limitations therefore system uncertainty always remains bounded.
Owing to the factor described above, bounded uncertainties and perturbations in the
elevation dynamics can be introduced as
Tg Tg Tg B1 B1 B1
a1 a1 a1

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)

where 1 K gyrou1 x6 cos x2 is azimuth affect on elevation and 2 k1 x7 - k2u1 is the

elevation affect on azimuth dynamics caused by the gyroscopic torque. Now by replacing
control laws (Eq. (5.32)) in Eq. (5.29) we get
c
V1 s1 ( x B1 x (a1 x ) b1 x Tg sin x2 1 K1sat ( s1 ))

(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)

From the bounds of the system states in Eq. (5.40)

0 x1 0.556

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.

5.2.2.1 Simulation Results

The controller designed in Eq.(5.32) delivers the following simulation results.
Figure 5-1 shows the regulation control of the helicopter output states. Figure 5-2 and
Figure 5-3 are having the phase portraits of elevation and azimuth respectively. The
controller effort to deliver the desired results is shown Figure 5-4. The sliding surface
designed to generate the control laws show their convergence in Figure 5-5.

___________________________________- 73 -_______________________________

Sliding Mode Control

0.7
Elevation
Azimuth
0.6

Angular Position (rad)

0.5

0.4

0.3

0.2

0.1

-0.1

5
Time (sec)

10

Figure 5-1: Regulation control of Helicopter outputs

Phase Portrait
0.05

Rate of Change of Elevation

-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

Figure 5-2: Phase Portrait for Elevation dynamics

___________________________________- 74 -_______________________________

Phase Portait II
0.1

Rate Of Change of Azimuth

-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

Figure 5-3: Phase portrait for Azimuth Dynamics

Sliding Mode Controller Effort
0.6
Elevation
Azimuth
0.4

Elevation

0.2

Controller Effort

Azimuth
-0.2

-0.4

-0.6

-0.8

-1

5
Time (sec)

10

Figure 5-4: Sliding mode controller effort for regulation

___________________________________- 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

Figure 5-5: Sliding manifolds convergence which ensures states convergence

5.2.2.2 Experimental Test Results

The implementation of the sliding mode controller designed in Eq. (5.32)
delivered the output states response as shown in Figure 5-6 and respective effort can be
seen in Figure 5-7. It can be observed that the equilibrium state in vertical plane is
achieved in 7 sec. The coupling in horizontal plane is introduced at 22 sec. It can be seen
that equilibrium position in horizontal plane is maintained by the controller with the
divergence of 1 degree for about 5 sec after the coupling has been introduced.
The controller guides the system both in vertical and horizontal planes to
equilibrium positions as shown in Figure 5-8 with effort shown in Figure 5-9. The results
show that the system when released at 40 deg. initially in horizontal plane, the SMC
controller governs the system, released in nonlinear range of horizontal plane, to
equilibrium point with in 5 sec with effort shown in Figure 5-9.

___________________________________- 76 -_______________________________

Sliding Mode Control

25

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 -_______________________________

Sliding Mode Control

25

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.

5.2.2.3 Performance evaluation

The performance indices for elevation dynamics in Table 5-1 and for azimuth
dynamics in Table 5-2 quantitatively certify the improved performance of the twin rotor
system dynamics with sliding mode control after the introduction of cross-coupling. The
sliding mode controller has improved performance as equilibrium position is well
maintained with lesser errors.

___________________________________- 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)

Table 5-1: Performance indices of Elevation Dynamics

Azimuth

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)

Table 5-2: Performance indices of Azimuth Dynamics

5.2.3 Sliding Surface Design via LMIs

Linear matrix inequalities (LMI) based sliding manifold has been proposed in
[32] The method described in [32] allows us to design a linear sliding manifold for twin
rotor system which assures the system performance and the control costs required to
maintain sliding, in an optimal way. The convex optimization problem has been
developed based on structural assumption on the Lyapunov matrix for the closed-loop
system. The solution of the convex problem allows us to attain optimized sliding surface
which helps in designing feedback controller.
Consider the system in Eq.(2.25) that can be reformulated as

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

A -0.6376 4.1157 14.5719 3.1223 -0.2015 -6.6450 -1.6767 (5.44)

0.5483 -0.1836 0.3187 0.8323 0.0061 0.2033 0.0748

-0.4637 2.8887 10.1974 4.9631 -0.2663 -8.6611 -1.1768

0
0
0
0
0
0
-3.3330

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)

The feedback controller L from sliding mode control can be designed as

L ( SB) 1 SA

(5.50)

Under given coordinate system, the switching function will be as

___________________________________- 80 -_______________________________

S S2 M

Im

(5.51)

Where
M m( n m )
S 2 mm

The partition of the system is now in the form of

x1 , x 2

associated with system

represented in canonical form.

We can now say, when
s0

(5.52)

x2 Mx1

(5.53)

x1 A11 x1 A12 x2 ( A11 A12 M ) x1

(5.54)

x2 A21 x1 A22 x2 Bu ( A11 A12 M ) x1 Bu

(5.55)

Thus

The new coordinate system can be introduced as

x ( I n A) x

(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)

The Eq.(5.63) can be reformulated in linear matrix inequalities as

) X X ( A BL
)T
( A BL

Qq X

R1/2 LX

XQqT
Iq
0

XLT R1/2

0 0
I m

(5.64)

And our minimization problem can be formulated as

Minimize Trace(( I n A)T X 1 ( I n A))

(5.65)

Subject to LMI in Eq.(5.64). The decision variables are L and X where

X
X 1
0

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)

The feedback controller based on sliding surface can be calculated as

___________________________________- 82 -_______________________________

L NX 1 [ N1 X 11

Im ]

(5.71)

M N1 X 11

(5.72)

And in original coordinates the feedback controller can be written as

L M

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

Angular Position (rad)

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

Figure 5-11: LMI based sliding mode controller effort

Sliding Surfaces
70
Elevation
Azimuth
60

Sliding Surface

50

40

30

20

10

5
Time (sec)

10

Figure 5-12: LMI based sliding surface convergence

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.

5.3 Higher order sliding mode control

Higher order sliding mode control [24], [25], [26] is a recent approach that allows
the removal of all standard sliding mode restrictions, while preserving the main sliding
___________________________________- 84 -_______________________________

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.

5.3.1 Super Twisting Algorithm

The 2-SMC super twisting algorithm [31] delivers control law for the sliding
surface with relative degree, r 1 . The continuous nature of the control law helps to
overcome chattering caused by sign function in sliding mode control. The designed
control law ensures the system convergence in finite time along with robustness against
the external disturbances and parameter uncertainties.
The controller design procedure first involves the reduction of twin rotor MIMO
system in two SISO systems and then two controllers can be designed for each system
based on their error dynamics defined as
E X X eq

(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)

Their respective time derivatives are

1
e2 1e2 0 e2

(5.77)

2
e5 3
e5 2 e5

(5.78)

The new local coordinates can be defined as

___________________________________- 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

From Eq. (2.18) to Eq. (2.21), we can say

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)

As the states are bounded and their range can be defined as

0 x1 0.556
0 x 0.4363
2

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

and 0 m1,2 M 1,2 such that

1,2 1,2 and

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

dynamics 2 0.62 , m2 28.40 , M 2 99.31 .

These bounds computation for helicopter model certify that there exist sliding
mode control and the 2-SMC super twisting controller can be designed for twin rotor
system.
The super twisting continuous control law [31] constitutes of two terms. The first
term is defined by means of its discontinuous time derivatives, while the other is
continuous function of the available sliding variable. Consider the control algorithm
ui (t ) ui '(t ) ui ''(t )

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 -_______________________________

5.3.1.1 Simulation Results

The simulations are carried out with the control laws in Eq.(5.89). The regulation
of the output states is shown in Figure 5-13 and their respective phase portrait is shown in
Figure 5-14. The crossings in the phase trajectories depict involvement of third state i.e.
motors state. The successful simulations validated the control laws for implementation on
actual experimental setup.
Output States Response
0.25
Elevation
Azimuth

Angular Position (rad)

0.2

0.15

0.1

0.05

10

15

Time (sec)

Figure 5-13: Regulation response of output states of dynamical model of helicopter

Phase Portrait
0.3
Elevation
Azimuth
0.2

Rate of Change of Angular Position

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

Figure 5-14: Phase portraits of elevation and azimuth dynamics

___________________________________- 88 -_______________________________

5.3.1.2 Experimental Results

The implementation of the controller in Eq.(5.89) with 1=1.5, 1=0.9 on
elevation dynamics delivered results shown in Figure 5-15. These parameters were
selected because system response was with lesser rise and settling time thus delivering
desired performance. The variation in CG is shown in Figure 5-16 which is serving as
uncertain parameter creating disturbance torque. The respective controller effort to
overcome the uncertainties and maintain the equilibrium position is shown in Figure
5-17. The sliding surface convergence can be seen in Figure 5-18. It can be observed that
the helicopter model acquires the referred position in 10 sec. and maintains the position
even with perturbed CG. In the first case, when the CG is perturbed 50% ahead of its
current value at 35 sec. the torque produced compels the helicopter model to hover with
nose bend down for 7 sec. and it deviates 9% from its referred path. When the CG attains
its normal position again the resulting torque makes the model to hover with nose tilted
up and with 7 sec. the controller nullifies the effect and attains the given path.
Elevation
25
Reference
Elevation

Angular Position (deg)

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

Figure 5-16: Variations in center of gravity serving as parametric uncertainty

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 -_______________________________

Sliding Surface Response

1

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

Figure 5-18: Elevation dynamics sliding manifold convergence

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

Angular Position (deg)

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

Figure 5-20: 2-SMC Controller effort to maintain equilibrium position in azimuth

___________________________________- 92 -_______________________________

Azimuth Sliding Surface

0.6

0.4

0.2

Sliding Surface

-0.2

-0.4

-0.6

-0.8

-1

10

20

30

40
Time (sec)

50

60

70

Figure 5-21: Azimuth dynamics sliding surface convergence

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

Nonlinear Rise Time Overshoots Settling

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

Table 6-1: Comparison between proposed controllers in this thesis

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 -_______________________________

7.1 MATLAB code of modeling of Helicopter Model

clc
clear all
close all

% System parameter values;

t1=0.3;
b1=0.00936;
I1=4.37e-3;
B1=1.84e-3;
Tg=3.83e-2;
t2=0.25;
b2=0.0294;
Tor=2.7;
Tpr=0.75;
Kr=0.00162;
I2=4.14e-3;
B2=8.69e-3;
Kg=0.015;
% System State Space Matrices
A=[-3.333 0 0 0 0 0 0;
0 0 1 0 0 0 0;
b1*1.667/I1 -Tg/I1 -B1/I1 0 0 0 0;
0 0 0 -1.3333 0 0 0;
0 0 0 0 -4 0 0;
0 0 0 0 0 0 1;
0 0 0 0.0899/I2 b2*2/I2 0 -B2/I2];
B=[2 0;
0 0;
0 0;
0.0625 0;
0 2;
0 0;
-0.0058/I2 0];
C=[0 1 0 0 0 0 0;
0 0 0 0 0 1 0];
D=[0 0;

___________________________________- 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];

7.2 MATLAB code for H with Traditional weights

model_linear
%%%%%%%%%%%%%Scaling
Win=[12 0;0 6];
G=series(Gss,Win);
T=feedback(G,eye(2));
S=1-T;
figure(2),sigma(G,{1e-3,1e3});
%%%%%%%%Weights%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

___________________________________- 100 -_______________________________

W1=01*[9000*tf(conv([1 2],[1 0.2]),conv([1

0.001],[1 1000]))

0;0

10000*tf(conv([1 2],[1 0.2]),conv([1 .001],[1 1000]))];

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
Gs=G*W1;
%%%%%%%%%%%%%%frequency alignment
Gf=freqresp(Gs,30);
Ka=align(Gf);
W1=W1*Ka;
Gs=G*W1;
% Conversion to Digital Domain
W1d=c2d(W1,0.01,'tustin');
W1ss=ss(W1);
W1ssd=ss(W1d);
[Wa,Wb,Wc,Wd]=ssdata(W1ss);
[Wad,Wbd,Wcd,Wdd]=ssdata(W1ssd);
figure(4),sigma(Gs,'r',G,'k',{1e-3,1e3});grid;
legend('Gs','G',1)
% Controller Designing
gammarel=1.1;
[Ks,gammin]=coprimeunc(Gs,gammarel);
Ks=-Ks;
Ksd=c2d(Ks,0.01,'tustin');
[Ac,Bc,Cc,Dc]=ssdata(Ks);
[Acd,Bcd,Ccd,Dcd]=ssdata(Ksd);
% For Implementation
Kco=series(W1,Ks);
Kcod=c2d(Kco,0.01,'tustin');
[Aco,Bco,Cco,Dco]=ssdata(Kco);
[Acdo,Bcdo,Ccdo,Dcdo]=ssdata(Kcod);
% Simulations
T2=feedback(series(G,Kco),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)');

___________________________________- 101 -_______________________________

7.3 MATLAB code for H with Hadamard weights

model_linear
% Scaling Weight
Win=[12 0;0 6];
% Hadamard Weighting of Scaling Weighting Function
G=Win.*G;
figure(2),sigma(G,{1e-3,1e3});
% Input Weighting Function
W1=01*[165*tf(conv([1 2],[1 0.2]),conv([1

0.001],[1 1000]))

0;0

160*tf(conv([1 2],[1 0.2]),conv([1 .001],[1 1000]))];

alpha=12;
s=2;
W1=W1.*[alpha s*alpha;s*alpha alpha];
% Hadamard Multiplying
Gs=G.*W1;
Gs=ss(Gs,'min');
figure(4),sigma(Gs,'r',G,'k',{1e-3,1e3});grid;
legend('Gs','G',1)
% Controller Designing
gammarel=1.1;
[Ks,gammin]=coprimeunc(Gs,gammarel);
Ks=-Ks;
Ksd=c2d(Ks,0.01,'tustin');
[Ac,Bc,Cc,Dc]=ssdata(Ks);
[Acd,Bcd,Ccd,Dcd]=ssdata(Ksd);
% Conversion to frequency domain
[NUM1c,DEN1c]=ss2tf(Ac,Bc,Cc,Dc,1);
[NUM2c,DEN2c]=ss2tf(Ac,Bc,Cc,Dc,2);
[r11c c11c]=size(NUM1c);
[r12c c12c]=size(DEN1c);
num11c=NUM1c(1,1:c11c);
den11c=DEN1c(1,1:c12c);

num21c=NUM1c(2,1:c11c);
den21c=DEN1c(1,1:c12c);

___________________________________- 102 -_______________________________

[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];

% Controller for Implementation

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)');

___________________________________- 103 -_______________________________

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

7.4 SIMULINK Diagram for H Implementation on Helicopter Model

Figure 7-1 SIMULINK Diagram for H Implementation on Helicopter Model

___________________________________- 104 -_______________________________

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

SIMULINK Diagram for H Implementation on Helicopter Model

Ele1

7.5

Figure 7-2 SIMULINK Diagram for H Implementation on Helicopter Model

___________________________________- 105 -_______________________________

x2

u1
x7

x2

u1

Elevation

x6

u2

u1

x7

x6

u2

Azimuth

Figure 7-3 SMC Controller Block

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

Figure 7-4 Elevation Controller Block

___________________________________- 106 -_______________________________

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

Figure 7-5 Azimuth Controller Block

___________________________________- 107 -_______________________________

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

7.6 SIMULINK Block Diagram for 2-SMC

Figure 7-6 SIMULINK Block Diagram for 2-SMC

___________________________________- 108 -_______________________________

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

Figure 7-7 2-SMC controller Block

___________________________________- 109 -_______________________________

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

[4]. Doyle, J.Grover, K.Khargonekar, and Francis,B. (1989),State space solutions to

standard H2 and H control problems. IEEE Trans. On automatic control, Vol 34.
no 8, pp831-846, August 1989
[5]. F.Van Diggelen and K. Glover, A Hadamard weighted loop shaping design
procedure for robust decoupling Automatica Vol. 30, No. 5, pp 831-845 1994
[6]. Slotine, J. J. and Li W. (1991), Applied Nonlinear Control, Prentice Hall, ISBN 013-040-890-5
[7]. Humusoft, CE 150 helicopter model: User's manual, Humusoft, Prague, 2002.
[8]. G. Karer and B. Zupani, Modeling and Identification of a laboratory helicopter.
Proceedings 5th MATHMOD Vienna, February 2006; (I.Troch, F.Breitenecker,
eds.); Vienna University of Technology; Vienna, Austria, 2006
[9]. M. Lopez-Martinez, F.R Rubio, Control of a laboratory helicopter using feedback
linearization European Control Conference (ECC-03), pp , 1-4 September,

Cambridge, UK, 2003.

[10]. M. Lopez-Martinez, J.M. Daz, M.G. Ortega and F.R. Rubio, Control of a
Laboratory Helicopter using Switched 2-step Feedback Linearization, Proc. of the

American Control Conf. (ACC04),2004.

[11]. M. Lopez-Martinez, F.R Rubio, Approximate feedback linearization of a
Laboratory Helicopter Sixth Portuguese Conference on Automatic Control

(CONTROLO 2004), Faro, Portugal, June 2004.

[12]. Te-Wei, Peng Wen, Time Optimal and robust control of Twin rotor system Proc.
of 2007 IEEE Int. Conf. on Control and Automation, Guangzhou, China, 2007.
[13]. Jun Yoneyama, Yukihisa Kikuchi, Robust control for uncertain systems with
application to helicopter model Proc. of the SICE 2002, Osaka.

___________________________________- 110 -_______________________________

[14]. M M. Lopez-Martinez, M.G. Ortega and F.R. Rubio, An H Controller of the

Twin Rotor Laboratory equipment, Proc. of 11th IEEE International Conference

on Emerging Technologies and Factory Automation (ETFA03), 2003

[15]. . M. Lopez-Martinez, C. Vivas, M. G. Ortega, A multivariable nonlinear H
controller for Laboratory helicopter. Proc. of 44th IEEE conference on Decision

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.

[19]. Skogestad, S. and Postlethwaite, I.(1996) Multivariable Feedback control

,Analysis and Design John Wiley & Sons. New York

[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

Decision and Control Maui, Hawaii USA, 2003

[22]. Tom Koudela, Renata Wagnerov, Position Control with Robust Algorithms
Proc. of the Portuguese conf. on Automatic Control, 2000.
[23]. Juhng-Perng Su Chi-Ying Liangt and Hung-Ming Chent, Robust Control of a
Class of Nonlinear System and Its Application to a Twin rotor MIMO System

IEEE ICITO2, Bangkok, Thailand.2002.

[24]. Levant A. Higher-order sliding modes, differentiation and output feedback control
International Journal of Control 2003; 76:924941
[25]. Levant A. Universal SISO sliding mode controllers with finite-time convergence.
IEEE Transactions on Automatic Control 1998; 46(9):14471451
[26]. S.V. Emelyanove, S.K. Korovin and A. Levant, Higher-order sliding modes in
control systems, Differential Equations, 29,pp. 16271647, 1993.

___________________________________- 111 -_______________________________

[27]. S. Laghrouche, M. Smaoui, X. Brun and F. Plestan, Second order sliding mode
controllers for pneumatic actuators, in American Control Conference ACC04,

Boston, Massachusetts, 2004.

[28]. Eds. K. Warwick,D. Rees, Multivariable Control:An Introduction To Decoupling
Control Industrial Digital Control Systems IEE Control Engineering series, Peter
Peregrinus 1988
[29]. Q. G. Wang, Decoupling control LNCIS 285,pp 115-128 2003
[30]. Richard C. Dorf, Robert H. Bishop, Modern Control Systems, 7th edition
Addison-Wesley Longman Publishing Co., Inc. Boston, MA, USA.
[31]. Wilfrid Perruquetti, Sliding Mode Control in Engineering, CRC Press, 1st edition
2002.
[32]. Christopher Edwards, A practical method for the design of sliding mode
controllers using linear matrix inequalities, Automatica, Volume 40, Issue 10,

October 2004, Pages 1761-1769

___________________________________- 112 -_______________________________