You are on page 1of 226

Rochester Institute of Technology

RIT Scholar Works


Theses Thesis/Dissertation Collections

2003

Aircraft landing gear simulation and control


Sanoopkumar Ambalaparambil

Follow this and additional works at: http://scholarworks.rit.edu/theses

Recommended Citation
Ambalaparambil, Sanoopkumar, "Aircraft landing gear simulation and control" (2003). Thesis. Rochester Institute of Technology.
Accessed from

This Thesis is brought to you for free and open access by the Thesis/Dissertation Collections at RIT Scholar Works. It has been accepted for inclusion
in Theses by an authorized administrator of RIT Scholar Works. For more information, please contact ritscholarworks@rit.edu.
AIRCRAFT LANDING GEAR SIMULATION AND CONTROL

by

SANOOPKUMAR AMBALAPARAMBIL

A Thesis Submitted in Partial Fulfillment of the Requirements for the Degree of

MASTER OF SCIENCE

IN

MECHANICAL ENGINEERING
Approved by:

Dr. Josef S. Torok


Department of Mechanical Engineering (Thesis Advisor)

Dr. Agamemnon Crassidis


Department of Mechanical Engineering

Dr. Benjamin Varela


Department of Mechanical Engineering

Dr. Edward Hensel

Department Head of Mechanical Engineering

DEPARTMENT OF MECHANICAL ENGINEERING


ROCHESTER INSTITUTE OF TECHNOLOGY
ROCHESTER, NEW YORK
NOVEMBER 2003
Permission Grant

I, SanoopKumar Ambalaparambil, hereby grant permission to the Wallace Memorial


Library of the Rochester Institute of Technology to reproduce my thesis in whole or part,
as long as no reproductions will be used for commercial use or profit.

Permission granted by:


SanoopKumar Ambalaparambil (Author of this thesis)

Date: '2.101; 101.


I
Acknowledgments

I would like to sincerely express my gratitude and thanks to Dr. Josef Tbrok for

giving me the excellent and challenging opportunity to work on a nonlinear system in

terms of both the dynamics and control aspects. I also thank him for his expert guidance

and motivation. I appreciate his great willingness to share knowledge and provide his best

support to the learning atmosphere.

This thesis work has been very fruitful for myself in regards to learning various aspects of
the nonlinear dynamics and the control design techniques. I also had the opportunity to
learn and utilize six softwares for my thesis work namely: Ideas for solid modeling of the

prototype model; Solidworks for creating graphical models; Maple for deriving the

equations of motion; Matlab for coding and toolboxes like sisotool and Itiviewer;
Simulink for running the simulation of the nonlinear model; and EXP for documentation

of my thesis work.

I would like to express my thanks to the faculty and staffs of Mechanical

Engineering Department at Rochester Institute of Technology for providing me the

necessary theoretical background and computer support in working on my thesis. I would

also like to thank my parents and my wife, Swapna for encouraging and supporting me

even though they are far away from me.

I also thank my friends at RIT who may have directly or indirectly influenced me
in writing my thesis. Last, but not the least I take this opportunity to thank God for always

leading me to the path of success and helping me bring the best of me in whatever I do.
Abstract

The focus of this work is the development of a Multi-degree Of Freedom (MOF)


model for simulation and control of the landing performances of the SAE Heavy Lift
Airplane model built by the Aerodynamics design team of RIT. From dynamic

considerations, the landing gear performance has two areas of interest: a) behavior during
touchdown impact, and b) response to excitation induced by track roughness during taxi,
take-off and later part of the landing runs.

As the problem at hand is highly nonlinear and complex, initially a simplified 1-

dof model of the same aircraft is derived and analyzed using Lagrange's equations and

energy methods. After understanding the basic dynamics of the system, a more complex

3-dof model is derived considering the dynamics of the wheels of the landing gear.

Analyses with and without the runway profile are carried out to study the effects of the

input signals on the derived model.

In order to achieve a controlled landing behavior of the aircraft, a linear damper or

shock absorber is incorporated into the analysis and simulation, even though it is not

available in the original prototype. The nonlinear models are linearized in Simulink and

control system designs are performed on the linearized models using Root Locus

techniques, Pole Placement methods, LQR approach, and Error Space approach. The

designed controllers are then applied on the original nonlinear 3-dof model of the landing
gear and a study of the effects of changing landing gear parameters is performed. Open

loop and Closed loop simulations are carried out to come up with an effective controller

type, which will ensure optimum landing performance in a normal landing situation.
List of Figures
Figure No: Title Page

Fig. 1.1 Landing Gear Types 16

Fig. 1.2 Heave-pitch Model 18

Fig. 3.1(a) Solid Model of the SAE Airplane (IDEAS) 29

Fig. 3.1(b) Kinematic Model (IDEAS) 30

Fig. 3.2 Analytical Model 31

Fig. 3.3 Load vs. Deflection of the Spring, ks 32

Fig. 3.4 Static Analysis 34

Fig. 3.5 F.B.Dof link EC 35

Fig. 4.1 Block Diagram of a Dynamic System 38

Fig. 4.2 Modeling Sequence 39

Fig. 4.3 One-degree of Freedom Model 41

Fig. 4.4 Plot of Potential Energy vs. 6 46

Fig. 4.5 Phase Curves 47

Fig. 4.6 Contour Plot of Total Energy 49

Fig. 4.7 3-D Plot of Total Energy 49

Fig. 4.8 Potential Energy Surface 50

Fig. 4.9 Implicit Plot of Total Energy 51

Fig. 4.10 Response vs. Time 53

Fig. 4.1 1(a) Generated Waveform and Phase Plot 54

Fig. 4.11(b) Frequency Contents Shown by FFT 55

Fig. 4.12(a) Simulink Block Diagram 56

Fig. 4.12(b) Response vs. Time 56

Fig. 4.13 Superimposed Plots 57

Fig. 4.14(a) Generated Waveform and Phase Plot 58

page 5
Fig. 4.14(b) Dominant Frequency Contents Shown by FFT 58

Fig. 4.15 FFT Comparison 59

Fig. 5.1 3-DOF Model 61

Fig. 5.2 Coordinates as a Function of 9 63

Fig. 5.3 Block Diagram Representation of the System of ODE's 67

Fig. 5.4 Responses of 6, z\, and z2 69

Fig. 5.5 Phase Plots 70

Fig. 5.6 3-D Plot of 0, zu and z2 71

Fig. 5.7 Results of FFT 72

Fig. 5.8 FFTof Z2 Shows Higher Frequencies 73

Fig. 5.9 Responses of the System with Runway Profile 76

Fig. 5.10 Phase Plot Representations of 0, z\, and z2 77

Fig. 5.1 1 Superimposed Responses 78

Fig. 5.12 FFT Outputs Plotted for First 75 Points 79

Fig. 5.13 FFTof Z2 Shows Higher Frequencies 80

Fig. 5.14 Superimposed FFT Plots 81

Fig. 6.1 Model with Damper 83

Fig. 6.2 Responses of the Damped System 86

Fig. 6.3 Damped vs. Undamped System 87

Fig. 6.4 Stages of Landing 88

Fig. 6.5 Landing With Nose Wheel Airborne 89

Fig. 6.6 Response at Touchdown 92

Fig. 6.7 Landing Simulation 96

Fig. 6.8 Landing Responses 97

Fig. 6.9 Energy Models 101

Fig. 6.10 Total Energy of the System During Landing 102

Fig. 7.1 Block Diagram of a Feedback Control System 103

page 6
Fig. 7.2 The Control System Design Process 104

Fig. 7.3 Block Diagram Representations of Open Loop


and Closed Loop Systems 105

Fig. 7.4 Definition of Rise Time, Settling Time, and Overshoot 107

Fig. 7.5 Regions in the s -


plane delineated by the transient
requirements: (a) rise time; (b) overshoot; (c) settling
time; and (d) composite of all three requirements 108

Fig. 7.6 Response of the Damped 1-DOF Nonlinear Model 112

Fig. 7.7 Linearization of the 1-DOF Nonlinear Model 1 13

Fig. 7.8 Nonlinear vs. Linearized Model 114

Fig. 7.9 Characteristics of the Linearized Model 115

Fig. 7.10 Transfer Function Model 118

Fig. 7.11 SISO Output of the PD Controller 118

Fig. 7.12 Step Response of PD Controller 119

Fig. 7.13 Transfer Function Model 120

Fig. 7.14 SISO Output of the Lead Controller 120

Fig. 7.15 Step Response of Lead Compensation Controller 121

Fig. 7.16 Transfer Function Model 122

Fig. 7.17 SISO Output of the PID Controller 123

Fig. 7.18 Step Response of PID Controller 123

Fig. 7.19 Comparison of PD, Lead, and PID Controllers 124

Fig. 7.20 Proportional Feedforward and Derivative Feedback 125

Fig. 7.21 Controller Performance on the Nonlinear System 125

Fig. 7.22 Lead Feedforward and Derivative Feedback 126

Fig. 7.23 Controller Performance on the Nonlinear System 126

Fig. 7.24 Proportional, Integral Feedforward, and Derivative Feedback 127

Fig. 7.25 Controller Performance on the Nonlinear System 127

Fig. 7.26 Comparison of the Three Controller Performances 128

page 7
Fig. 8.1 Static Equilibrium of the Plane 133

Fig. 8.2 Linearization of the Nonlinear 3-dof Model 135

Fig. 8.3 Nonlinear vs. Linear Responses 137

Fig. 8.4 Root Locus Plots of the Transfer Function Models 139

Fig. 8.5 Block Diagram Representation of Full State Feedback 143

Fig. 8.6 Root Locus Plot of the Desired Pole Locations 146

Fig. 8.7 Step Response of the Chosen Pole Locations 147

Fig. 8.8 Linear Responses of the Pole Placement Design 151

Fig. 8.9 Linear Responses of the LQR Design 155

Fig. 8.10 Simulink Diagram to Implement LQR Design 156

Fig. 8.1 1 Cost Function and the Integrand 157

Fig. 8.12 Linear Simulation Results of the LQR Integral Control 159

Fig. 8.13 Feedback Control with the Estimator 162

Fig. 8.14 Simulink Diagram to Design and Estimator 164

Fig. 8.15 Comparison of the Actual and the Estimated States 164

Fig. 8.16 Simulink Diagram of LQR Controller with Integral Feedback 166

Fig. 8.17 Controller Performance on Nonlinear and Linear Models 167

Fig. 8.18 Input Signals Generated for 1 sec 171

Fig. 8.19 LQR Error Space vs. LQR Integral One State Feedback 177

Fig. 8.20 Implementation of LQR Error Space Approach on Nonlinear Model . 180

Fig. 8.21 Control of the Nonlinear 3-dof Model 180

Fig. 8.22 Linear vs. Nonlinear Responses 181

Fig. 8.23 Responses Obtained After Adding x = 0.031 182

Fig. 8.24 Plot of the State Variables x, x$, and xq 183

Fig. 8.25 Control Efforts 183

Fig. 8 .26 Total Energy Dissipated During the Control Action 184

Fig. 8.27 Controller Behavior in Various Initial Conditions 185

page 8
Fig. 8.28 Controller Performance under Various Values of km and kn 186

Fig. 8.29 Controller Performance under Various Values of ci and c2 187

Fig. 8.30 Simulation with Different Values of v0 188

Fig. 8.31 Simulink Block Diagrams of the Actual Landing Scenario 196

Fig. 8.32 Landing Responses in Touchdown and Taxiing 197

Fig. 8.33 Responses of 0, i\, and z2 198

Fig. B.l Comparison of the Responses 222

page 9
List of Tables
Table No: Title Page

Table 3.1 Details of the Plane 28

Table 3.2 Inertia Parameters 30

Table 3.3 Linkage Dimensions 31

Table 3.4 Spring Constants 31

Table 3.5 Load vs. Deflection Data 31

Table 5.1 Block Diagram Representation of the Subsystems 68

Table 8.1 Derivatives of the Input Signals 172

page 10
Nomenclature
Notation Description Unit

(...) First Derivative w.r.t Time

(...) Second Derivative w.r.t Time

o Amplitude of the Runway Profile m

A, B, C, D Plant Coefficient Matrices

ith
Acceleration of the Particle
m/s2

a(s) Denominator Polynomial

A',B',C',... Coordinates Representing the Displaced Position of the Model


a Constant Value of the Pitch Angle During Landing deg

b(s) Numerator Polynomial

BD Length of the Link BD m

Constant Nose Wheel Linear Damper


m/s2

c Damping of the

C(s) Laplace Transform of the Controller

cx,Cy,cz Damping Constants N s/m

D Rayleigh's Dissipation Function Joule

d Distance Traveled Before the Nose Wheel Touches the Runway m

Time
^ Total Derivative w.r.t

Partial Derivative w.r.t Time


J^
DC Length of the Link DC m

St Virtual Displacement m

As Deflection of the Spring, ks N/m

E(s) Laplace Transform of the Error Signal, e

ED Length of the Link ED m

Etot Total Energy of the Plane Joule

ith
F Force on the Particle N

page 11
/ Frequency in Hertz Hz

Fd Force Acting at the Point D TV

FFT Fast Fourier Transform

Fs Force Due to the Spring, ks N

Fy Forces in y-direction JV

g Acceleration Due to Gravity


m/s2

G(s) Plant Transfer Function

H(s) Sensor Transfer Function

hnc Center of the Nose Wheel From the Static Equilibrium Position m

i,j,k Summing Indices

Ig Rotational Mass Moment of Inertia kg-m?

K Feedback Gain Matrix

kr),kp,ki Proportionality Constants for Derivative, Proportional, and Integral

Controllers

KE Total Kinetic Energy Joule

km Spring Constant of the Main Wheel N/m

kn Spring Constant of the Nose Wheel N/m

ks Spring Constant of the Suspension Spring N/m

L Lagrangian Joule

I Wavelength of the Runway Profile m

A Inverse of the Wavelength for a Full Rotation by 27r


m~l

Ls Unstretched Length of the Spring, ks m

m Number of Inputs

ith
rrii Mass of the Particle kg
M Mass of the Airplane kg

Md Moments About Point D N m

MF Moments About Point F N m

page 12
Mm Mass of the Main Wheel kg

Mn Mass of the Nose Wheel kg

p Vector of Control in Error Space

%MP Percentage Overshoot

n Degrees of Freedom

np Number of Poles

nz Number of Zeros

N Input Scaling Factor


ith
Pi Linear Momentum of the Particle

p Pole of the Transfer Function

PD Length of the Link PD m

PE Total Potential Energy Joule

Pm Pitching Moment N m

<f> Angle Between Links BD and DC deg


ith
qi Generalized Coordinate of the Particle

Q Weighting Matrix for the States

QF Length of the Link QF m

QG Length of the Link QG m

Qk Generalized Force N

Q*k Generalized Force Not Derivable From a Potential Function N

QP Length of the Link QP m

Laplace Transforms of the Reference Input, r


R(s)
R Weighting Matrix for the Control Inputs

ith
r,x,y,z Physical Coordinate of the Particle

r Reference Input

r.
m
Radius of the Main Wheel m

Fin Ground Reaction at the Main Wheel Point N

page 13
rn Radius of the Nose Wheel m

Rn Ground Reaction at the Nose Wheel Point N

s Laplace Variable

a The Real Part of the Pole

T Total Kinetic Energy Joule

t Variable Representing Time sec

T(s) Transfer Function of the System

r Time Constant of a First order System sec

9 Pitch Angle of the Plane deg

6(s) Laplace Transform of theta

6e Equilibrium Point of 9 deg

9t Angle of the Asymptote with the Real Axis deg

tp Peak Time sec

tr Rise Time sec

ts Settling Time sec

T Modal Transformation Matrix

u Vector of Inputs

u(t) Unit Step Function Heaviside Function

Ugrav Potential Energy Due to Gravity Joule

Uks Potential Energy Due to the Spring, ks Joule

Usp Potential Energy Due to the Springs Joule

V Scalar Potential Function Joule

vq Velocity of the Plane During Landing m/s

vq,vq,vf Velocities of the Points Q, G, F etc m/s

W Weight of the Plane Acting at the Center of Gravity N

u> Circular Frequency rad/s

uja Approximate Frequency of the Signal z# rad/s

page 14
cud Damped Natural Frequency rad/s

u>n Natural Frequency rad/s

xq,xh ^-coordinates of the Points J and H From the Origin m

xe (t) Equilibrium Conditions on the State Variables

Xi(t) Initial Conditions on the State Variables

x\,x2,...xn Plant State Variables

xi Integral State Variable

x(t) Estimate of the Actual State

x(t) Error Between the Actual and the Estimated State

Vector of Error State Space

y Vector of Outputs

Y(s) Laplace Transform of the Output, y

z Zero of the Transfer Function

zq,zh Vertical Coordinates of the Runway Profile at Points J and H

z\ Vertical Displacement of the Center of the Main Wheel m

z2 Vertical Displacement of the Center of the Nose Wheel m

C, Damping Ratio

page 15
Chapter 1

INTRODUCTION

1.1 Overview of
Landing Gears

One of the many critical components that make an aircraft function is the landing gear.

The basic function of a landing gear on an aircraft is to maneuver it during its ground

operations which include taxi, takeoff and landing. Of these, the most critical phase is the

landing because it involves a massive amount of energy transfer and the system has to be

stable enough to operate under these conditions. Some of the constraints and

requirements for the performance of landing gears are crash survivability, riding
performance, weight, height, stroke length, retraction, and steering. All of the above

would have to be optimized for the reasonable performance of an aircraft on its ground

operations.

Although there are so many different variants of the landing gear models, the

conventional one has a tired wheel unit, a shock absorbing unit and a supporting structure

[1]. The wheel unit has a main wheel assembly attached to the fuselage and a nose wheel

assembly attached to the nose of the aircraft. There are three common types of landing
gear: conventional, tricycle, and tandem. This work has been confined to the tricycle type

of landing gear. During landing, the main wheels come in contact with the ground first

which is called the touchdown and then the nose wheel makes contact with the ground

[2]-

There are three commonly adopted arrangements for the nose gear geometry which are

telescopic, articulated, and semi articulated as shown in Fig. 1.1 (Conway, 1958).

\\v\ \ \ \ \\y\\\\ \\\\\ w

Telescopic Articulated Semi articulated

Fig. 1 1.
Landing gear types

page 16
Chapter 1 Introduction

The telescopic design is kinematically simpler but the geometry may not account for the
large strut forces. Since the shock strut in the articulated design has both ends hinged, it
produces only axial forces and so does minimize the journal frictional forces, but at the

cost of complex kinematics. The shock strut in the semi articulated gear design has only
one end hinged, but the performance can be optimized by proper selection of gear

parameters.

The two major concerns for landing gear performance are: (a) behavior during touch

down impact, (b) and performance excitation induced by track roughness during taxi,
takeoff and the later part of landing runs. The high level of transients induced during
touchdown have to be controlled smoothly in order to achieve the steady state in a

reasonable amount of time.

1.2 Developments in the Field of Landing Gears

There has been considerable research in the field of landing gear design and development

and most of these works are specific to a particular case under consideration. The various

steps in landing gear design, its performance requirements and shock strut design are well

described by Currey [3], based on experience and experiments. Jocelyn [4] of NASA has

explained in detail the landing gear dynamics, especially shimmy and brake-induced

vibration and has well summarized the work documented from the last ten years to

highlight the latest efforts in solving these vibration problems. According to their work,

includes and brake-


the landing gear vibration self-induced oscillations called shimmy

induced vibration. Possible causes for shimmy are low torsional stiffness, excessive

freeplay in the gear, wheel imbalance, or worn parts. Brake-induced vibration includes

gear walk, squeal and chatter, which are caused by the frictional characteristics between

the brake rotating and nonrotating parts. This paper also explains the work done by
Moreland [5] in landing gear dynamics and the theory of shimmy. Moreland found that

to precisely describe the system and the shimmy phenomena, the mathematical model

damper-
required 5 degrees of freedom: tire deflection, swivel angle, strut deflection,
linkage strain, and airframe motion. Even though it may seem cumbersome to include

these degrees of freedom to the specific model of study, they play an important role if the

model has to be comprehensive enough to consider all the practical working conditions.

W. Kriiger et al. [6] documented the aircraft landing gear dynamics in a wider sense,

considering both landing and ground maneuvering with control aspects. There are two

control concepts which are of interest to automotive systems: (a) in a fully-active

page 17
Chapter 1 Introduction

suspension an actuator provides force which is directly defined by a control law, (b) in a

semi-active suspension only the damping force in the direction of the momentary damper
displacement is modulated to control the damping coefficient. Their work analyzed the

problem from design requirements to the simulation of the system with all probable

variables.

Most of the earlier models were simplified to reduce the complexity of the resulting

equations due to lack of computational facilities and they assumed the gear geometry to

be telescopic. D. Yadav and R. P. Ramamoorthy [2] analyzed an extended version of the

heave model created by Yadav and Kapadia (1990). This model idealized the aircraft as a

rigid beam supported by the shock absorbers over two wheel springs. The masses of the

wheels and the linkages were approximated as lumped at the respective wheel axes and

the sprung masses were assumed to have heave and pitch degrees of freedom.

The heave model neglected the pitch degrees of freedom and its coupling with the model,

whereas the heave pitch model incorporated these effects into the analysis. The heave

pitch model used a two-wheeled articulated nose gear and telescopic main gear

geometries during landing impact. They also studied the influence of linkage dynamics

with oleo pneumatic shock struts for the first time. Their model is shown in Fig. 1 .2.

Fig. 1 .2 Heave-pitch model [2]

In this study, three shock absorber parameters -


orifice discharge coefficient (Cd), initial
air pressure (Pa0) and pneumatic area (Aa) were selected for variation in the articulated

nose gear. By varying Cd for the nose gear, the system behavior indicated that the

touchdown should take place with a smaller damping (larger Cd)- This would not have

major effects on the maximum heave or pitch amplitudes of the sprung mass. They also

page 18
Chapter I Introduction

summarized a smaller initial air pressure at touchdown, that could be made to increase

with time for the impact duration, could be used in active control of shock absorbers.

1.3 Objectives of Current Work

1 . The focus of this work is to develop a Multi-Degree of Freedom (MDF) model for
simulation and control of the landing performance of the SAE Heavy Lift
Airplane model built by the Aerodynamics Design Team of RIT.

2. Initially, a simplified model of the same system is studied and the equations of

motion are derived. In this particular model, the equations are derived

considering only the pitch angle of the plane and the wheels are approximated

with massless points which move on the road surface. This leads to a single-

degree-of-freedom system which has the basic characteristics of the system

behavior. This simplified model also allows study of the energy equations and

their variation due to the pitch angle.

3. A Lagrangian formulation and energy methods are used throughout the study to

derive the equations of motion.

4. Both models, with and without the runway profile are studied and analyzed to

understand the effect of the road perturbations on the model. FFT analyses are

performed to find out the frequency content in the response signals.

5. The existing model has only a spring on the nose wheel, which induces oscillatory
behavior to the response. In order to achieve a controlled response behavior, the

system is analyzed with a linear damper or shock absorber to dissipate the

dynamics during the touchdown impact.

6. A Root Locus study is performed on the characteristic equations of the linearized

models to study the effects of changing gear parameters on the stability of the

landing gear. By analyzing the pole-zero placements of the characteristic

equation, it is possible to understand the effects of the characteristic roots on the

response behavior.

7. Because the model derived is highly nonlinear in nature, initially three types of

controllers are designed for the linearized version of the 1-dof model using root

page 19
Chapter I Introduction

locus techniques and Matlab sisotool. These controllers are then modified to suit

the needs of the nonlinear model.

8. Open loop and Closed loop simulations are carried out to come up with an

effective controller type, which will ensure optimum landing performance in a

normal landing situation.

9. A discussion of the actual landing scenario is presented where the main wheels

contact and ride on the runway until the landing speed is lowered by the pilot. The

actual landing simulation is carried out in Simulink, considering both these cases.

10. A more sophisticated approach is followed in designing a controller for the 3-dof

model. The 3-dof model is linearized in Simulink and the linearized state space

model is used in designing the feedback gains and ensuring minimum steady state

error to a step input.

11. The linear controller is applied to the nonlinear 3-dof model and modified to suit

the design requirements of the overall landing situation.

12. To precisely describe the system, parameters like tire deflection, swivel angle,

strut deflection, damper-linkage strain, and airframe motion will have to be

incorporated into the model as a recommendation for future work.

page 20
Chapter 2

LAGRANGE'S EQUATIONS OF MOTION

2.1 Introduction

The derivation of Newton's equations of motion require that a mechanical system

consisting of several components be broken apart into different elements and the forces
and moments acting on the components be identified on the free-body diagram of each

element. The dynamic equations are formed based upon the forces and moments acting
that specific In
on element of the system. case of complicated multi-body systems, it is
advantageous to derive the equations of motion from a global perspective without

worrying about the forces at the interconnections. J. L. Lagrange (French mathematician,

1736-1813) came up with an effective method of


deriving the equations of motion from

system energy considerations. Since energy is a scalar quantity, this method does not

require the often confusing sign conventions in assuming the force directions on the

freebody diagrams.

2.2 Lagrangian Formulation

Lagrangian mechanics uses generalized coordinates of a system, qi, instead of the

physical coordinates r^ [7]. This generalized approach enables one to derive the
equations of motion which are independent of any particular coordinate system or set of

generalized coordinates.

Consider a system of N particles with n degrees of freedom and we define a set of

generalized coordinates qi, i =


1, . . .
n, which has a transformation from the physical

coordinates.

Now for the z'-th particle the governing equation of motion in vector form is:

m^i =
i (2.1)

or:

t =
F- (2-2>

where p^ is the linear momentum of the z-th particle, which is given by:

page 21
Chapter 2 Theory ofLagrange 's Equations ofMotion

Pi =
rrniri (2.3)

Now let us find how these equations transform under the transformation from physical to
generalized coordinates. The time rate of change of the generalized momentum

corresponding to the k-th generalized coordinate qk is given by:

d .

,
d ( dT ,

where Tis the total kinetic energy of the system with respect to the physical coordinates

which is given by:

N
1

Now generalized momentum p^ can be written as:

AT
dT ir-^
( di dy\ dz
J^TOil ^i^+^i^ +
.

^i^T1
Pit =
jrr =
] (2.6)
5gfc f V dqk dqk dqk

From chain rule we have:

dx{ ^ dxt . dx{

Taking partial derivatives of the above equation with respect to the generalized velocity

qk results in the following relation:

dx-
dXi
m
dqk dqk

Using this relation Eqn. (2.6) becomes:

page 22
Chapter 2 Theory ofLagrange's Equations ofMotion

N
dT
mi\xi^ f . dxi
+ yi~
. dy{ . dzi
+ Zi~
(2-9)
dqk *ldqk ldqk

Taking the total time derivative of Eqn.


(2.9) and applying the product rule to the terms in
the summation results in:

N
d (dT dxi dy{ dzi
Y^mi
.. .. ..
x,.
=
+
'
yt
+ Zl (2-10)
dt \dqk 9qk atdqk
i=i d^k
N '

( dxi d . .
d fdy{ d ( dzi
Xi\ + Zi\
-
-

=i dt\dqkJ yidt\dqk) %dt\dq

By Newton's Second Law we have:

rrnyi =
Fiy
I*
m>iZ i iz

The first summation term in Eqn. (2.10) then becomes:

N N
.. dx{ .. dy{ .. dzt d* dy^_
E
i=l
rrti
.
Xi-

oqk
hy^-

dqk
^zi~y-

dqk E
i=l dqk ydqk
F.dzz

dqk

in which the right hand side is the generalized force Qk represented by the transformation
equations. The second summation can be interpreted as follows:

Eqn. (2.7) can be written as:

N
d(xi) ^\dxi .
dx
=

dt ^dq^j+~dt

Taking partial derivative of a;, with respect to qk in Eqn. (2.7) we get:

page 23
Chapter 2 Theory ofLagrange 's Equations ofMotion

d ( dxi Ed2Xi d2Xi


dt \dqk 3
j-(dqjdqk dtdqk
d i^ydxi . dxi

dxt
dqk

Therefore the time rate of change of the k-th generalized momentum is given by:

N
d
Vk~
-
(dT\ Qk + y^m Xi-
di .

\-yi~ dy{ V z{
. dz{
(2-11)
dt\dqk) 1=1 9qk ldqk '"dqk
d
Qk
y2

+ + +
dqk zty
dT
=
Qk + k =
1, 2, ... ,n

dqk

Eqn. (2.11) relates the generalized force Q^to the time rate of change generalized

momentum. Hence the equations of motion of the system in generalized coordinates qk

are:

d ( dT\ dT
=
Qk k =
l,2,...,n (2.12)
dt\dqk) dqk

Eqn. (2.12) is known as the general form of Lagrange's Equations of Motion. There is

one equation corresponding to each generalized coordinate qk. It can be seen that the

dynamics of the system is characterized by the kinetic energy and the virtual work done

by the generalized forces.

For a conservative system, there exists a scalar potential function in terms of the

generalized coordinates:

V =
V(quq2,...,qn)

The generalized forces can be derived from this potential function as:

page 24
Chapter 2 Theory ofLagrange 's Equations ofMotion

dV
Qk=
(2-13)
-

dqk

Since the potential function only depends on the generalized coordinates not the

velocities,

dV

dqk

so we have:

dT __
d(T -

V)
(2-14)
dqk dqk

Substituting the generalized force and Eqn. (2.14) into the Lagrange's Equations (2.12),
we obtain:

d_ d(T-V) d(T-V) = 0
dt dqk dqk

The scalar quantity in the parentheses is defined as the Lagrangian function:

L(q,q,t)=T(q,q,t)-V(q)

Thus it can be seen that the Lagrangian is a function of the generalized coordinates and

velocities. It represents the difference between the total kinetic energy and the total

potential energy of a conservative system. Now the equations of motion (2.12) can be
written as:

dL
d_( dL (2-15)
dt \dq\ dqk

It can be summarized that for a conservative system, all the dynamics are characterized by
a single scalar function, the Lagrangian of the system.

page 25
Chapter 2 Theory ofLagrange's Equations ofMotion

2.3 Dissipative Systems

Forces which can not be derived from a potential function are called nonconservative

forces. For example, forces associated with friction cannot be conservative, due to
dissipation of energy. Consider a system with resistive damping forces, with components

proportional to the velocities of the particles. For such a system we have:

CxiXi
F CyiVi
F
J- IT

The virtual work done by these dissipative forces under a set of virtual displacements <5r

is:

sw =
Yy
i=l
1
Sr

^2(cxiXi8xi + Cyiyfiyi + cziZi8zi)


i=l
N r n
dx{ dyi dzi
-
j=l Lfc=i dq dqk dqk W
E M^+^^+c^/d
n N
ri d ( \
/ J o / ., a
^
\ cxixi * Cyiyi > czizi
J %
k=l L i=l

Therefore the generalized forces associated with the dissipative forces are:

QI [C-xixi CyiVi czizi


*~

J
~

~T
/ a
*hd<ikK
.

Cyiyi CziZi I
a .
/ A C-xyEj r ~~r

dq\i-x

We can define a scalar function of the generalized velocities given by:

N
1 ( 2 2\
D + Cyiyi cziZi I
=
y ( cxjXj
^
~r

i=i

page 26
Chapter 2 Theory ofLagrange 's Equations ofMotion

Substituting back into the equation for virtual work, the dissipative generalized forces are

derivable from the function D as:

6W =
YlQnkc8qk
fc=l

tldu

where D is known as the Rayleigh's Dissipation Function. The most general form of

Lagrange's equations of motion can thus be written as:

d (dL\ 9L dD

in which L = T V is the Lagrangian, D is the dissipation function, and


Q*k is a

generalized force not derivable from a potential function or a dissipation function. We

can note that Rayleigh's dissipation function represents one-half the rate at which energy
is being dissipated or in another words average loss of power in a nonconservative

system.

page 27
Chapter 3

MODEL DESCRIPTION AND STATIC ANALYSIS

3.1 Introduction of the Model

3.1.1 Model Specifications

The model used in this study is the SAE Heavy Lift Airplane built by the Aerodynamics
Design Team of RIT. It tricycle
uses a
landing gear geometry with a nose wheel

operating on a spring and two cable reinforced main wheels. Table 3.1 shows the details
of the airplane. For modeling and analysis purposes, the airplane body can be modeled as

a rigid body with mass concentrated at the center of gravity of the plane. The landing
gear model, which is of interest here, can be modeled as massless linkages consisting of a

nose wheel mounted on a and two main wheels.


spring

Max Payload 14.5 lbs


Engine Brake HP 1.9 HP@ 16, 000 rpm
Airfoil Modified SeligA310
Thickness 14.5 % of Chord
Camber 4.5 % of Chord
Wingspan 112.5 in
Chord 8.5 in
Aspect Ratio 12.5
Wing Control Surfaces 24 in Flaperons

Wing Construction Fiberglass and Balsa


Shear Skin Carbon Fiber
and Balsa Sper
in3
Cargo Capacity 350
Fuselage Construction Carbon Fiber and Balsa
Tail Construction Balsa Truss and MonoKote
Main Landing Gear Cable Reinforced Dubro Gear
Nose Landing Gear 20 lb Dual Strut Nose Gear
Receiver 6 Channel Futaba FP -

R127PF
Servos Tower Hobbies TS -

53 STD

Table 3.1 Details of the Plane

page 28
Chapter 3 Model Description and Static Analysis

\ ! i

Fig. 3.1(a) Solid Model of the SAE Airplane (IDEAS)

6
Fig. 3.1(b) Kinematic Model (IDEAS)

Fig. 3.2 shows the analytical model of the SAE Heavy Lift plane.

page 29
Chapter 3 Model Description and Static Analysis

Fig. 3.2 Analytical Model

3.1.2 Data Measured From the Model

To completely describe the system and analyze the response behavior, it is necessary to
have a reasonable amount of data of the actual system. The following Tables (Tables 3.2,
3.3, and 3.4) show the actual data measured from the physical system shown in Fig. 3.2.

It is to be noted that the data maybe subjected to measurement errors.

Parameter Magnitude Unit


M 11.84 kg
kg-m2
Ig 1.966

Mm 0.021 kg
Mn 0.031 kg
Tm 0.064262 m

rn 0.033655 m

Ls 0.03 m

Table 3.2 Inertia Parameters

page 30
Chapter 3 Model Description and Static Analysis

Linkage dimensions:

Parameter Magnitude Unit


QF 0.165 m

QG 0.0127 m

QP 0.3048 m

DC 0.035 m

PD 0.132 m

ED 0.046 m

BD 0.063 m

Table 3.3 Linkage Dimensions

Spring Constants:

Parameter Magnitude Unit


k
""m 8000 N/m
kn 8000 N/m
Ks 5676.7088 N/m

Table 3.4 Spring Constants

Here km and kn are the spring constants of the main and nose wheels respectively. For the

analysis purpose, a high stiffness value of 8000 N/m is assumed for both the springs. ks,
which is the stiffness of the nose wheel mounting spring, is derived by performing the
load deflection test and plotting the Load vs. Deflection graph (Fig. 3.3). Table 3.5

shows the data of the test.

Fs (lb.) Deflection (in)


0 0
1 0.022
2 0.072
3 0.096
5 0.15
6 0.18
7 0.22
8 0.24

Table 3.5 Load vs. Deflection Data

page 31
Chapter 3 Model Description and Static Analysis

Load Vs Deflection

Deflection (inches)

Fig. 3.3 Load vs. Deflection of the Spring, ks

From the graph, the equation describing the line is:

y =
32.415 x

Slope of the line gives the spring constant, ks = 32.415 lbs /in which is equivalent to

5676.7088 N/m.

3.2 Assumptions

1 . The analysis is performed assuming a two dimensional model. The effects of the road

perturbations across the width of the road are neglected.

the center of gravity.


2. The airplane body is modeled as rigid, with mass concentrated at

All the connecting elements are assumed to be massless linkages.

3. The tires are modeled as masses attached to linear springs of high stiffness values

because the prototype model has tires which have aluminum wheels and rubber rings

landing, the main wheels contact the runway


wound on them. During an actual since

page 32
Chapter 3 Model Description and Static Analysis

first, a linear damper is included with the main wheel spring in order to make the

landing smoother.

4. In deriving the mathematical model of the system, it is assumed that all the forces are

two dimensional.

5. Since the Lagrangian formulation considers only those forces which contribute to the

total energy of the system, the reactions at the interconnections are not considered in

the analysis.

6. The road surface is modeled as a sine wave with small amplitude and the airplane

velocity during landing is assumed to be 10 m/s.

7. The friction forces at the joints and on the road surfaces are not considered in the

analysis.

8. Taylor series expansions of order 3 are performed on the equations of potential

energy, and zh to cancel out the higher order terms which would make the equations

highly nonlinear.

9. An approximation of coscfi is used in deriving the equations of motion for the 3-dof
model.

10. The suspension is modified with a linear damper; where as the actual system may

have oleo-pneumatic shock absorber which is nonlinear.

1 1 Actuator
. and sensor dynamics are not considered in designing the controllers for the

system.

page 33
Chapter 3 Model Description and Static Analysis

3.3 Static Analysis

In order to analyze the equilibrium value of the angle </> of the link CE, a static analysis

of the model is performed [8]. This model is a simplified approximation of the original

one in the sense that it approximates the wheels as massless points which move on the

road surface. When the system is in equilibrium, the forces acting are the reaction forces

from the ground and the weight of the plane as shown in Fig. 3.4. Since there is no

motion, the effect of friction on the road surface is neglected. Now we can apply the

equilibrium conditions to find out these reaction forces and the angle </> when the plane is

in static equilibrium.

Fig. 3.4 Static Analysis

Summing all the forces in y-direction yields:

0
^Fy =
0^Rm + Rn-W =

Substituting for the weight of the plane we get:

Rm + Rn = 116.1504 N (3-1)

Summing all the moments about the point F:

J^Mp =
0^Rn-(QP-ED-
sincj)) -W-QG = 0

Simplifying for i^we get,

page 34
Chapter 3 Model Description and Static Analysis

W-QG
Rn =
(3.2)
(QP -

ED sin<j))

Considering the equilibrium of the link CE (Fig. 3.5) we have:

Fig. 3.5 F.B.D of link EC

Taking moments about D we have:

^Md = 0 => -Rn-DE-


sine/) + Fs DC sin<\> =
0

ED
S~DC'^

Substituting for Rn we get:

ED W-QG
F =
(3.3)
DC (QP -

ED sine/))

Since the deflection of the spring ks is restricted by the points A and C we also have:

Fs = ks(BD -

DC cos(f)
-

Ls) (3.4)

Substituting for all known parameters and equating Eqns. (3.2) and (3.3) we get:

1.938
= 5676.7088(0.033 -

0.035 cos<f))
(0.3048 -

0.046 sin(f>)

If we solve the above equation for 0, we can see that there are two real and two imaginary
roots. Since imaginary and negative values for (j) are not practically possible we have:

page 35
Chapter 3 Model Description and Static Analysis

4> =
24.67 degrees

This is the value of the angle <f> when the system is in static equilibrium. Now we can

find the static deflection of the spring ks from Eqn. (3.3) as:

BD-DC cos(f> -Ls = 0.0012 m

We can also calculate the deflection of the spring ks from the load vs. deflection graph

(Fig. 3.3) by calculating the spring force in Eqn. (3.2). By substituting the value of ^ we

have:

Fs =
6.786 N, which is equal to 1.525 lbs.

The equation of the line is:

y
=
32.415 x

Here y = 1.525 lbs and solving for x we get:

x =
0.00119 m

which is very close to the static deflection value obtained by the static analysis.

Now the reactions from the ground can be found from Eqn. (3.2):

Rn = 5.165 N

Rm =
110.985 N

The above analysis shows that the assumptions made are reasonable and the formula for

the spring force is a good approximation of the actual situation.

page 36
Chapter 4

MODELING OF THE SYSTEM

4.1 Dynamic Systems

4.1.1 Introduction

This chapter focuses on


deriving the equations of motion of the system under

consideration. Although we can represent the dynamics of a physical system in several

ways, the analysis is performed with equations which are compatible with the

mathematical and numerical methods to The importance


simplify the solution methods. of

deriving equations of motion for a system is manifested by the fact that the dynamics of a

system can be characterized by the governing differential equations of motion. Unlike the
static analysis where the response of the system is independent of time, the dynamic
analysis involves characterizing the behavior of the system which responds to input

signals, disturbance signals, and initial conditions.

4.1.2 Definitions Related to Dynamic Systems

Before we proceed to derive the equations of motion, it is helpful to know some of the

definitions related to a dynamic system. The definitions of a dynamic system [9] are

described below.

A system is a set of interacting components connected together in such a way that the

variation or response in the state of one component affects the state of the others.

Modeling is the process of identifying the principal physical dynamic effects to be


considered in analyzing a system, writing the differential and algebraic equations from the
conservation laws and property laws of the relevant discipline, and reducing the equations

to a convenient differential equation form.

The major disciplines of engineering systems are mechanics, electricity and electronics,

fluid mechanics and fluid controls (including hydraulics and pneumatics), and

thermodynamics .

The behavior of a system is characterized by its response to external inputs, disturbances,


and initial conditions. Fig. 4.1 shows the simple description of a dynamic system. Here
outputs mean the dependent variables of the differential equation that represent the

page 37
Chapter 4 Modeling of the System

response of the system. Inputs mean functions of the independent variable of the
differential equation, the excitation, the function to the External
or
forcing system.

disturbances or perturbations mean those external environmental effects that may occur

randomly or unexpectedly. The Initial conditions are the initial values of the dynamic
variables of the system. The dynamic variables of a system are those variables whose

time derivatives appear in the governing equations.

Initial Conditions

Inputs Outputs
Dynamic System

Disturbances

Fig. 4.1 Block Diagram of a Dynamic System

A dynamic system is described by differential equations and the response of the system

is determined by the present state of the system (the initial conditions) and the present

input. Thus, a dynamic system may continue to have a time-varying response after the

inputs are held constant.

The transient response of a dynamic system to an external input refers to the behavior of

the system as it makes a transition from the initial to the final condition. A dynamic

system will reach a steady state after all of the transients have died out. The time it takes

to reach the steady state is called the settling time.

A differential equation describing the characteristics of a dynamic system will have

dependent and independent variables, their derivatives and excitation or external forces.
In general, differential equations can be either linear or nonlinear. A linear differential

equation consists of a linear combination of the system variables and their derivatives.
On the other hand, a nonlinear differential equation has nonlinear combinations of the

variables and their derivatives. Typically, these equations may contain the product of two

variables, square of a variable, trigonometric functions of the variables, and so on. There
are numerous methods available for solving linear differential equations, but most

nonlinear systems do not have known analytic solutions. If an analytic solution is not

possible for a nonlinear system, we find an approximate solution by numerical integration

page 38
Chapter 4 Modeling of the System

methods which are


normally done by digital simulation. These simulations are run with

small time increments and the solutions are typically approximations of the actual

solutions.

4.2 Modeling of Dynamic Systems

The following Fig. 4.2 describes the various steps in modeling a dynamic system.

l -* Original Design
ification
Actual Physical
Dynamic System

-
to Be Considered
Modeler's Perception
of Dynamic System

-i Write Component and

System Equations
Mathematical
Representation

Classical Differential Equations


*~
\ 1 ransfer t unction
State-Space Equations
Calculated
Response

Analytic Solution
1 m
Digital Simulation

Performance
Analog Simulation
Analysis

Static Gain
1
Disturbance Sensitivity
Dynamic Characteristics

Fig. 4.2 Modeling Sequence

The first step is identifying the actual dynamic system of interest. It has all the inherent
dynamic response characteristics of the system that correspond to the exact linear or

nonlinear behavior of the system. The actual system has the true response which sets the

goal of the analysis.

Second is the engineer's perception of the system. For ease of modeling and analysis,
the modeler may neglect some of the nonlinearities or higher order dynamic

page 39
Chapter 4 Modeling of the System

characteristics; however, the actual system is a true representation of all of these effects

and characteristics. Therefore, the engineer's perception might not be a true

representation of the actual dynamic system.

The third stage, which is the mathematical model of the system represented by the

differential equations derived using physical laws and principles. If the system is
energy
nonlinear, the mathematical representation could include some approximations to

simplify the analysis; therefore, the resulting equations may just be an approximation to

the engineer's perception of the real system.

In the fifth stage, the calculated response obtained by analytical solutions to the

governing differential equations of the system is an exact solution to those equations.

However, errors might exist between the response obtained by numerical or simulation

methods and the actual solution of the differential equation.

The fifth stage is the analysis of the performance of the dynamic system. Several
analytical methods could be used to analyze the performance of the system such as

frequency domain, time domain analysis, and stability analysis.

One of the most important steps is identification of the parameters which influence the
performance characteristics of the response of the system, and the adjustment of these

parameters until an acceptable performance is achieved. Thus the feedback and iterative

step in Fig. 4.2, in which the system is modified, is an important part of the engineering

process.

It should be noted that since the calculated response using numerical simulation is the

solution to the approximate mathematical model which was derived using simplifying

assumptions. It is three steps removed from the response of the actual system. This
means that the solutions to the governing differential equations might not be an exact

representation of the actual system of study; however, the trends in the response of the

system could be understood by doing a simulation. By looking at the response, one can

identify the parameters which influence these trends. Then by adjusting these parameters,
and making the appropriate modifications to the model and simulation, a reasonable

performance can be achieved.

page 40
Chapter 4 Modeling of the System

4.3 Derivation of One Degree of Freedom Model

4.3 1. Coordinate System

The degrees of freedom of a system are the number of independent coordinates necessary
to describe the motion of the system at any instant of time. So as the name suggests a

One-degree Of Freedom system requires


only one coordinate to describe the motion at

any instant of time. Initially, a simplified model is derived to find the response

characteristics and energy behavior of the system. In this particular model, the wheels are

approximated as massless points which slide on the road surface. So the degree of

freedom has been simplified to the pitch angle of the plane which describes the angular

motion of the plane with respect to time. Let 9 be the pitch angle of the system. The

following Fig. 4.3 shows the graphical model of the one degree of freedom system.

///////y ///////////
xD

*4

Fig. 4.3 One-degree of Freedom Model

Here xq and zq are the x and y coordinates of the point F with respect to a rectangular

coordinate system referred to as global coordinate system. These are basically x, y, z

coordinates with respect to an axis triad with fixed unit base vectors i , j, k or e"x, e"y, e"2.

However, in many cases it becomes useful to define other coordinate systems referred to

as a local coordinate systems with their origin fixed at the current position of a point. In

page 41
Chapter 4 Modeling of the System

our case, 0is defined as the angular motion about the center of gravity of the plane, which

is a local coordinate.

4.3.2 Velocity and Energy Analysis

Since the point F is tracing the road profile defined by the vertical coordinate zq, we can

write its vertical velocity as z'o, which is the time derivative of zq . Now the velocity of

point Q can be written as:

~$q =^f + -nr-^ 9 QF (4.1)

We have the velocity of the center of gravity as:

-V>G=^Q+ N0-QG (4.2)

Substituting for v q in Eqn. (4.2), we get:

-$G =^F + 9-QF+ N 0 QG


e^^

1?G =t0 + zVf + <-0 QF + 10 QF + <-0 QG + 1* QG

of the of the center of gravity, we obtain:


Separating the x and y components velocity

9-QG- sin9
~vgz cos9
-
-9-QF-

=^o

T0-6-QF- + 9 QG
~vcy = sin9 cos9

and:

v2G
=
v2Gx + v2Gy (4.3)

The total kinetic energy of the system consists of both translational and rotational kinetic

energies. So we have:

page 42
Chapter 4 Modeling of the System

1-Ig(9)2
KE=Im.v2g +
2
(AA)

Substituting for v2G we get:

KE= (v0-9-QF-cos9-
-M

2 9-QG-sin9) (4.5)
"1
1
.n

+ (io -9-QF-
sin9 + 9 QG +
-7G(
0
j
The total potential energy of the plane is due to gravity and the spring ks. Therefore, we

have:

i Hj U grav ~> U
sp

where Ugrav is the potential energy of the plane due to the y coordinate of the center of

gravity of the plane yG and is given by:

QF- cos9
U^av =
M-g-(yF + + QG sin9) (4.6)

and
Usp is the spring potential energy due to the deflection As of the spring ks and is
given by:

USP=\-
ksA2s (4.7)

Substituting for As = BD -

DC cos<f> -

Ls in Eqn. (4.7), we get:

Usp=\ ks(BD -

DC cos(f> (4.8)

Therefore, the total potential energy becomes:

PE = M-
g(z0 + QF cos9 + QG sin9) (4.9)
Ls)2

+ - -

ks(BD-DC cos(j)
-

page 43
Chapter 4 Modeling of the System

where 0 is a function of 0. To find 4>, we have:

(QF '
cos9 + Qp sine ~

PD cos9)
cos{6 -

9)) -

(4.10)
ED

Taking inverse cosine of the above equation, we obtain:

(QF '
cosd +p '
sin0 ~

PD '

((h-e) =
cos-1

cos6\
,X(QF cosO + QP sin9
-

PD cos9
4> = 0+
ED

Now taking cosine of the above equation, we get:

_i
/ QF cos9 + QP sin9
-

PD cos9
COS(j) cos 9+ (4-11)
ED

Expanding out the above identity:

.
( QF cos9 + QP sin9 -

PD cos9\
f (4.12)
J
cos(f> = cos9

QP-sin9-PD-cos9^'J
,'QF-cos9 +
-szn0M/l I
^

This expression for cos<p can be substituted in Eqn. (4.9) to find the total potential energy
of the system. By analyzing the equations for kinetic energy and potential energy, we can

see that the kinetic energy is a function of the angular position coordinate 0, its derivative
0, and the independent variable t. Therefore, we can write KE /(0, 9 ,f). The

potential energy is a function of 0 and time, that is PE =


g(9, t). Now we can find the

total energy of the system which is a very important criterion in the analysis of a dynamic

system. Thus we have:

Etot = KE + PE

page 44
Chapter 4 Modeling of the System

Etot = \m (v0-9-QF-cos9-
9-QG-sin9) (4.13)
)*

+ (i0 -9-QF-
sin9 + 9 QG + -IG( 9

M-g-(z0 + QF- cos9


+ QG sin9) + \
Zt
ks(BD -

DC coscj) -
Ls)2

Since there are no dissipative components in the above equation, it represents a

conservative system, which satisfies the law of conservation of energy. That is, energy
can neither be created nor destroyed, but can change from one form to another. Hence,
the total energy of the system is independent of position. It is an invariant of the motion

and its value depends only on the initial conditions. A conservative system is one for
which all working forces are derivable from a potential function.

The potential energy, which is a scalar function can be graphically represented by a curve
or a surface (depending on the number of variables). If we draw horizontal lines parallel

to the axis representing the independent variables, we can collect points which have the

same value of potential energy corresponding to each line. These are called level sets,

which are defined as locations of constant potential energy in the physical domain of the

system. These level sets represent equipotential points, curves or surfaces depending
upon the number of variables.

In order to visualize the potential energy for the system of study, we first assume the road

perturbations are zero, that is z$ = 0. Substituting this in the equation for PE, we get:

,2
M-g- (QF DC
PE = cos9 + QG -

sin9) +
-

ks(BD -

cos(f> -

Since the expanded form of the potential energy contains so many nonlinear terms, a

Taylor Series expansion of order 3 is carried out to come up with a simplified form for

the potential energy. By substituting for the parameters (please refer Appendix A.l for

energy derivation), we get the following equation:

02
PE = 19.3415 -

7.8215 0 + 103.0882 (4.14)

The above equation represents a curve in the physical domain of the system and is plotted

in the following Fig. 4.4.

page 45
Chapter 4 Modeling of the System

Potential Energy vs. Theta


38-

-0.3 -0.2 DA D.2 D.3


Theta

Fig. 4.4 Plot of Potential Energy vs. 0

The horizontal lines represent level sets of total energy of the system for different initial

conditions. Since the potential energy can not exceed the total energy the allowable

motion of the system is restricted to regions in which:

PE <E

For a particular total energy level, the difference between total energy and the potential

energy is the kinetic energy. At the boundary where PE =


E, the kinetic energy must be
zero. These are called turning points where the displacement is maximum and the

system is momentarily at rest and it changes direction. At the locations where the

potential energy is a relative minimum as compared to the kinetic energy, the dynamics is

described by a potential energy well.

But within the specified limit of total energy, depending on the initial conditions the

potential energy and kinetic energy counterbalance each other.

Substituting for the known parameters in Eqn. (4.13), we get the total energy as:

page 46
Chapter 4 Modeling of the System

R'tot 0.1621 + 0.1624


cos29 sz'n20 + 0.983 W (4.15)
+ 19.3415 -

7.8215 9 + 103.0882 02

The above equation represents an implicit relationship between the position of the

particle, 0, and the velocity of the particle, 0. It defines a family of closed curves in a

plane with axes denoted by 0 and 0. This equation is plotted in the following Fig. 4.5, for
different values of the parameter Rtot.

Fig. 4.5 Phase Curves

Each individual curve of velocity versus displacement is called a phase curve and it

shows the simultaneous evolution of the position and velocity of the particle. The

collection of all these curves associated with various values of the total energy is known
as the phase plane.

The locations where the velocity and acceleration of the particle are zero are called the

equilibrium points. In other words, these are locations of the physical domain where

the net force acting on the particle is zero. These points are found by setting:

d(PE) 0= -7.8215 + 206.1764-0


d9

page 47
Chapter 4 Modeling of the System

Solving this, we get 0 =


0.0379 rad and the corresponding value of potential energy is
given by, PE = 19.1932 Joules. These locations correspond to stationary values of

potential energy, that is infinitesimal changes in position cause no variation of the

potential energy. Since the net force is zero at these points, we have static equilibrium at

these locations. The location of these points are on the 0-axis because at an equilibrium

point the velocity is equal to zero.

The location of these equilibrium points on the potential energy curve helps us to

determine the stability of the system, that is how the system will respond to small

perturbations from the state of equilibrium. For a conservative system, if the equilibrium

point, say 0e lies at a local minimum of the potential energy then the resultant force on the

particle is an attracting force, where as if the equilibrium point lies at a local maximum,

then the resultant force is repelling. In other words, a physical system always tries to

achieve a position of lowest potential energy. So an equilibrium point 0e is stable if:

d92

which means, the curve proceeds with a positive curvature at the point 0e.

and is unstable if:

^^<0d92

or the curve has a negative curvature just after passing the point 0e.

d
If ^P =
0, then we need to do further analysis to find the stability of the system.

Now (Fig. 4.6) the total energy, we get concentric loops


doing a contour plot of

associated with different values of energies.

page 48
Chapter 4 Modeling of the System

-0.4

Fig. 4.6 Contour Plot of Total Energy

A 3-dimensional plot (Fig. 4.7) of the total energy with respect to 6 and 9 can be done to

see the variation of energy more clearly.

thdot

Fig. 4.7 3-D Plot of Total Energy

page 49
Chapter 4 Modeling of the System

It is to be noted that the above analysis was performed by assuming zero velocity of the

plane (v0 =
0) and zero road perturbations or runway profile (z0 =
0). So the
displacement of the system is due to the gravitational force at the center of gravity since

there is no other external force acting on the system and the nose wheel mounting spring
induces an oscillatory motion with small amplitudes.

If we incorporate the plane velocity and the runway profile into the analysis the energy

equations are modified as follows:

Let the runway profile be of the form:


'

2 7T Vq
zq a sin l t (4.16)

where a is the amplitude and I is the wavelength of the runway profile, vq, represents the

velocity of the plane during landing. Accordingly, the new potential energy becomes:

92
PE =
19.3415 + 0.2904 sin(20 n t)
-

7.8215 9 + 103.0882 (4.17)

Now the potential energy is a function of 9 and t and it describes a surface. As the time

varies the potential energy curve follows the sinusoidal pattern along the time axis. The

following graph (Fig. 4.8) shows this variation.

Fig. 4.8 Potential Energy Surface

It can be seen that the kinetic energy is now modified with contributions from the plane

page 50
Chapter 4 Modeling of the System

velocity, vq and the velocity due to road perturbations, z0. So we get the new kinetic

energy as:

KE = 592 -

9 (19.536 cos9 + 1.504 sin9) + 0.146 cos2(62.83 t) (4.18)


+ 1.145(0) +9 cos(62.83 t) 0.0236 cos9 -

0.3068 sin9

the above that there is the kinetic


By looking at equation we can see a component of

energy which is constant and is determined by the plane velocity. The total energy is

also modified with contributions from the plane velocity, vo, and the velocity due to road

perturbations, zq, and is given by:

Etot =
0.146 cos2(62.83 t)
-

9 (19.536 cos9 + 1.504 sin9) (4.19)


+
1.145(0)2

+ 9 cos(62.83 -

1) [o.0236 cos9 -

0.3068 sin9

92
+ 0.2904 szn(62.83 t)
-

7.8215 9 + 103.088 + 611.3415

For a set of initial conditions, the energy is a fixed value and it describes a phase curve

on a plane of 0 and 9 (phase plane) and the phase curve follows the sinusoidal pattern

This be clearly seen from the following Fig. 4.9, which shows
along the time axis. can

the energy surface plotted for a specific set of initial conditions.

thdot 12

Fig. 4.9 Implicit Plot of Total Energy

page 51
Chapter 4 Modeling of the System

By analyzing the system in terms of its energy aspects, we have seen that the total energy

is a fixed quantity for a given set of initial conditions. This represents a conservative

system, for which all the forces are derivable from a potential function. We have also

seen that there is always a balance between the potential energy and the kinetic energy of

the system, as any one of these can not exceed the total energy as imposed by the law of

conservation of energy.

4.3.3 Equation of Motion for 1-DOF Model

Now we are ready to derive the equation of motion of the system based on the kinetic

energy and potential energy. The difference between the kinetic energy and the potential

energy of the system is called the Lagrangian, which is given by:

L = KE-PE (4.20)

By Substitution we have:

9QG-
l=\m (uo -9-QF-

cos9
-

(4.21)
2

+ (z0 -

0 QF sin9 + 0 QG- +
H)
2

M-g-(z0 + QF- cos9 + QG sin9) +


-

ks(BD -

DC cos<j> -

Lsf

The equation of motion is given by Lagrange's Equation:

dL(d_L\_d_L
=

dt\d9j d9

that the and the road perturbations are equal to zero, we obtain
Assuming plane velocity
the degree of freedom system as (please refer
the equation of motion describing single

Appendix A.2 for derivation):

0 + 90.0234 0 = 3.415 (4.23)

describing the dynamics of


This is the single second order ordinary differential equation

model with no tires and zero plane velocity and road perturbations.
the simplified

page 52
Chapter 4 Modeling of the System

Equation (4.23) can be integrated to find the response of the system as time varies.

Taking Laplace Transform of eqn. (4.23), assuming zero initial conditions, we get:

(s2 3.415
+ 90.0234) 9(s) =

or:

3.415
Q(s) s(s2 (4.24)
+ 90.0234)

Taking the Inverse Laplace Transform of the above equation we get:

9(t) =
0.0379 1 -

cos(9.488 t) u(t) (4.25)

where u(i) is the unit step or Heaviside function which ensures that the response starts

when t = 0. Eqn. (4.25) represents a sinusoidal response and is plotted in the following
graph (Fig. 4.10).

Response vs. Time

Fig. 4.10 Response vs. Time

From the graph, it can be seen that if the initial conditions on 0 and 0 are equal to zero,

page 53
Chapter 4 Modeling of the System

then the response starts with 0 =


0 and proceeds
sinusoidally with a maximum amplitude

of 0.0758 rad, which is equal to 4.34 deg.

Running an FFT (please refer Appendix A.3 for FFT codes) on the signal is useful in

finding the frequency content of the signal. For this purpose, the simulation is run to

generate 145 data points which give a reasonable amount of information about the

dominant frequencies contained in the signal. It is apparent that, the response signal has
two dominant frequencies; one at 0 Hz and the second one at 1.51 Hz. This is calculated

as follows:

In fact, the original signal can be written as:

0(f) =
0.0379 cos(2 tt 0 t)
-

0.0379 cos(2 tt 1.51 t)

So by ranning an FFT we expect two peaks where the amplitudes are collected; one at 0

Hz and the other at 1.51 Hz. This has been verified by the output of the FFT algorithm

(Fig. 4.1 1(a) and 4. 11(b)).

Generated waveform for 145 points

6 6
Time (sec)
Plot ofThdot vs. Theta
Q.4

0.2

-0.2

-0.4

j 0.01 0.02 0.03 D.04 0.05 O.OB 0.07 O.OB


Theta

Fig. 4.1 1 (a) Generated Waveform and Phase Plot

page 54
Chapter 4 Modeling of the System

Magnitude of Theta plotted against Hz

4 -

o
2 --

L^ i i ~i+

Magnitude of Theta plotted against Hz

"

2
i i j | j
3
Hz

Fig. 4.1 1 (b) Frequency Contents Shown by FFT

As expected, the FFT output shows that the dominant frequencies are 0 Hz and 1.51 Hz.

We can also see that, there are minor frequencies of very small amplitudes present in the

signal as shown by the non-zero band between OHzand 1.51 Hz, which maybe due to

noise present in the signal.

Now let us find the response of the system modified with the plane velocity and the road

perturbations. Introducing the plane velocity and the runway profile (Eqn. 4.16) into
equation (4.22) we get:

9 + 90.0234 0 =
3.415 + i-0(0.853 sin9 -

0.0656 cos9) (4.26)

The above differential equation can be numerically integrated in Simulink to find the

response. Results are shown in the following figures (Fig. 4.12(a) and 4.12(b)).

page 55
Chapter 4 Modeling of the System

Analysis with road perturbations

80 0234

Galn2
thdotl

To Woriepacel
?
-+F
?
Scope
5eope1

Integrator Integrator! To Woikspace

.a53-sin(uI1D-0
005e"co^u[1D

1"pi"pi"sinr20'pi"(u[1EO -J

0-

To Wortepace2

Fig. 4.12 (a) Simulink Block Diagram

Response vs. Time

Time (sec)

Fig. 4.12(b) Response vs. Time

page 56
Chapter 4 Modeling of the System

This modified response looks very similar to the response without the plane velocity and

road perturbations, the reason being that the assumed road perturbations signal is of very
small amplitude and thus it does not cause major changes to the behavior of the system.

However, it is interesting to note the differences by superimposing the two responses as

shown in the following Figure 4.13.

0.08

0.5 1.5 2 2.5 3 3.5 4.5


Time (sec)

0.4 I 1 -i 1 1 i

: :
0.2 -

Response with Road Profile


i
j Response without Road Profile
2.
.

to
0
\-

-0.2

i i
r=-
1 _
i i
-0.4

0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.03


0
Theta

Fig. 4.13 Superimposed Plots

in the phase plot represents 9, it can be noticed that there are


Since the vertical axis

as compared to the graph with no perturbations. This


relative variations of velocity
in kinetic energy and potential energy as the total energy is
signifies the relative changes

now a function of the road perturbations, as shown by the energy plots.

Now let us run the FFT on the response signal to find the dominant frequencies present.

in the (Fig. 4.14(a) and 4.14(b)).


The results are shown following graphs

page 57
Chapter 4 Modeling of the System

Generated waveform for N1 points

Fig. 4.14(a) Generated Waveform and Phase Plot

Ma gnitude of Theta plotted against Hz


a 1 1 1

BC

TO
ai A
H
jz
V-

C>
2

0 TSffijfflSfiSffiSrff KflS*
C1 1 2 3 4 5 B 7
H7
Magnitude of Theta plotted against Hz
B l 1 1 1

6
ro

| 4

2 Lj
n a " J _i 1 1

0 1 3 4
Hz

Fig. 4.14(b) Dominant Frequency Contents Shown by FFT

page 58
Chapter 4 Modeling of the System

As earlier, the dominant frequencies of the parent signal are still at 0 Hz and 1.51 Hz.
But we can see more minor frequencies present between 0 Hz and 1.51 Hz and after

1.51 if 2. Notice the frequency 2.92 #2, which has been introduced as a result of the

road perturbations signal. A comparison of the FFT outputs is shown in the following
Fig. 4.15.

Fig. 4.15 FFT Comparison

The qualitative analysis (energy methods) and the equations of motion, both describe the

inherent dynamic properties of the single degree of freedom system. Now that we have a

clear picture of the basic dynamics of the system, we can proceed further to include other

degrees of freedom.

page 59
Chapter 5

MULTI-DEGREE OF FREEDOM MODEL

5.1 Introduction

We have seen the system behavior with respect to one of the most important degrees of

freedom, that is the pitch angle. In order to analyze the system with its actual behavior,
we have to include the effects of tire dynamics as well. If we incorporate the tires into the
original model, it introduces two more degrees of freedom, namely the displacements of

the center of the wheels. In a practical sense, the tires of an airplane


play a very important
role during landing, taxi, and later parts of the landing run. Ned J Lindsley and Nitin B.
Talekar [10], in their tire model have described the tasks in deriving and discretizing the

equations of motion for the tire. Their other areas of interest included developing an

empirical model for the tire's inflation characteristics, loading and rolling of the tire in the

vertical plane, imposing out-of-plane deflections. Results of tire interaction were

analyzed with a landing gear model using ADAMS. The study of tires itself is a different
area of study.

The airplane in our present analysis makes use of aluminum wheels which have rubber

rings wrapped around them. As a good approximation, they can be modeled as lumped

masses with linear springs. The spring constants of these wheels are assumed to be high,
of the order of 8000 N/m. No dampers are used in this tire model because the wheels

have very little damping.

5.2 Equations of Motion

5.2.1 Velocity Analysis

The nose and main wheels are modeled as spring mass systems with masses Mand

Mm and with spring constants Kn and Km, respectively. Let rn and rm be the radii of the

nose and main landing gears and z\ and z2 be the displacements of the springs Km and

Kn, from the static equilibrium position of the system. The static equilibrium position

of a system is that position in which the weight of the system is balanced by the static

spring forces. Let (x0, z0) and (xh, zh) be the coordinates of the points J and H from

the origin of the global coordinate system. Let 0 be the pitch angle of the plane, and <f> the

angle between links BD and DC. Fig. 5.1 shows the geometric representation of the

model with tires added.

page 60
Chapter 5 Multi-degree ofFreedom Model

/////// S\S ////////// / /


xO

*H

Fig. 5.1 3-DOF Model

Assuming the displacements ziand z2 are vertical, the coordinates of the center of gravity

(Please refer Fig. 5.2) are given by:

xG xq QF sin9 + QG cos9 (5.1)

yG =
ZQ + Zl + QF cos9 + QG sin9 (5.2)

derivatives of these position coordinates, we get the velocity components


Taking the time
of the center of gravity as:

vgx =
vo-9-QF-
cos9
-

9 QG sin9 (5.3)

vGy =z0 + z1-9-QF-


sin9 + 0 QG cos9 (5.4)

Therefore, the velocity of the center of gravity is:

page 61
Chapter 5 Multi-degree ofFreedom Model

2
\lVGx + v\Gy
=
vg (5.5)

5.2.2 Energy Derivation

We notice that the total kinetic


energy now has contributions from Mm and Mn.
Therefore we have:

lMm(zi)2 l-Mn(z2)2
KE=\M-#G+ \lG(ef + + (5.6)

Substituting for vG from equation (5.5), we get:

KE = \u {vq-9-QF- cos9 - 9QG-


sin9) (5.7)

+ (i0 + ii -

0 QF sin9 + 9 QG

\lG{Q)2 \Mm(zy)2 )-Mn(z2)2


+ + +

As before, the potential energy has two components: (1) due to gravity, and (2) due to

spring forces. The potential energy due to gravity can be written as:

Ugrav =
9 M(zQ + 21 + QF cos9 + QG sin9) + Mm-Zl + Mn-z2 (5.8)

Before we find the potential energy due to the springs, we need to derive an expression

for cos(j) in terms of the coordinates 0, z\, and z2. Let us refer to the following Fig. 5.2.

page 62
Chapter 5 Multi-degree ofFreedom Model

~}z}_ S^cEqwUbrium Position f

Fig. 5.2 Coordinates as a Function of 0

In the above figure, it can be seen that the link DC makes an angle of (<j> 0) with the

vertical axis. So cos((j>


9) can be written as:

(21 + QF cos9 + QP sin9 -

PD cos9 -

z2)
cos((j) 0) =

ED
(5.9)

Taking inverse cosine of both sides of the above equation, we obtain:

_i (z\ + QF cos9 + QP sin9 -

PD cos9 -

z2
(4> 0) = cos

+ QF- cos9 + QP sin9 -

PD cos9 -

., ,'zi z2
4> = 9+ cos
ED

Taking the cosine of both sides of the above equation, we get:

-i fzi + QF cos9 + QP sin9


-

PD cos9
-

z2
COS(j) = cos 9+ cos (5.10)
[ ED

Expanding out the above identity:

page 63
Chapter 5 Multi-degree ofFreedom Model

QF-
zi + cos9 + QP sin9 -

PD cos9 -

z2
coscf) =
cos9
ED (5.11)

21 + QF cos9 + QP sin9 -

PD cos9 -

z2
szn0 \ /1 -

ED

As found earlier, the potential energy of the spring, ks, is given by:

Ls)2

Uk3 =
ks(BD -

DC coscf) -

(5.12)
g

If we substitute for cos<f) in equation (5.12), we obtain an elaborate form for the potential

energy. Moreover, in deriving the equations of motion, we are required to evaluate the

partial derivative

afl
09
=
ks(BD -

DC cos<t> -

LS)DC sin<f>
^
d9

where:

ifzi + QF cos9 + QP sin9 -

PD cos9 -

z2
sincj) = sin 9+
ED

and

QF- cos9
d4 d_\ -ifzi
+ + QP sin9
-

PD cos9
-

z2
0+
d9 d9 ED

In order to avoid cumbersome expressions, small displacements are assumed. Therefore

set

'

zx + QF cos9 + QP sin9
-

PD cos9
-

z2
coscf)
ED
(5.13)

The displacements of the end points of the springs km and kn are 20, 21 and 2# , z2

respectively. Hence their deflections are given by (21 20) and (z2 zjf). Now the total

potential energy due to all the springs is given by the following expression:

page 64
Chapter 5 Multi-degree ofFreedom Model

Ls)2
Usp= ks(BD DC coscj)
(5.14)
- - -

Zq)2

+ '

km{zi ~

+ -

kn(z2 -

ZrY
^

Hence, the total potential energy is:

PE =
g M(z0 + 21 + QF cos9 + QG sin9) + Mm 21 + Mn z2 (5.15)
1 9 1 9 1
+ -

ks(BD -

DC coscj)
-

Lay + -

km(zi -z0y + --i

A Taylor series expansion of order 3 is carried out on the potential energy to come up
with a simplified form.

The Lagrangian of the system is given by:

L = KE-PE (5.16)
9-QG-
L=-M (v0 -

9 QF cos9 -

sin9)
2

+ (z0 + zi-9-QF sin9 + 0 QG

\Mm(Zl)2 \Mn(z2)2
+ \Ig{0? + +

QF- cos9
M(z0 + zt + + QG sin9) + Mm-z1 + Mn-z2
1 . , s2
1 , . ,.. .
^2

ks(BD -

DC coscj)
-

LaY -

^2 km(zx -

z0)
--

kn(z2 -

zH)
2

5.2.3 Lagrange's Equations of Motion

There are three equations of motion, one each for 9, zx, and z2. These are given by the
Lagrange's Equations as:

page 65
Chapter 5 Multi-degree ofFreedom Model

d_(dL\ dL
dt\de)~"de~

d_(dL\ dL
~
=

dt \dTj ~dz\
dL
( \

dt \ dz2 J dz2

Initially, let us assume that there are no road perturbations


(20 =
zH =
0) and no plane

velocity (v0 = 0). By evaluating the terms in the above set of equations and
substituting
for the data, we arrive at the following three equations of motion (please refer Appendix
B.l for the derivation of the Equations of Motion):

Equation of Motion 1:

0 21 (0.853 0.065 cos9) + 21 (47.353 0


=
szn0
437.37) (5.17)
-
-

+ 22(437.37 47.353 0) + (3.892 125.43 0)


-
-

Equation of Motion 2:

z\ =
0(0.1647 sin9 -

0.013 cos9) + (277.07 z2


-

951.55 21) (5.18)


(0)2
02
+ 0.1647 cos9 + 0.013 sine] + (4.57 -

84.45 0 -

6.94)

Equation of Motion 3:

22 = (106012.07 21
-

364076.59 z2
-

1109.31) (5.19)
+ (32312.48 0 -

1749.2 02)

The above three equations are the mathematical representation of the dynamic behavior of

the 3-DOF model with tires added. By looking at these equations, we can see that they
are dynamically coupled in the three coordinates 0, z\, and z2. In other words, the
differential equation for 0 has terms containing 21 and 22 and vice versa. There are no

direct integration methods to solve this system of nonlinear ODE's. So we can use

approximate solutions calculated at definite time steps called numerical methods.

Reasonably accurate numerical methods for solving a system of nonlinear ODE's are

known as Runge-Kutta Methods. Of the many available Runge-Kutta Methods, the

page 66
Chapter 5 Multi-degree ofFreedom Model

classicFourth-Order Runge-Kutta Method has been found to be


very efficient (very good
accuracy) for solving the system of ODE's (please refer Appendix B.2 for Matlab codes to
solve the 3-dof model using ode-45 function). However, the same system of ODE's can

be analyzed using block diagrams in Simulink. Fig. 5.3 and Table 5.1 show the block
diagram representations of the above system of ODE's in Simulink. The numerical

integration has been performed using the ODE-45 solver.

Ti WdIspi2
TtWotapml

z1do12

th -

A. ~z2dot

aj

^T
ToWotap

?
ndot ToWotspace4

Simulation of the system with no mad

perturbations and no plane velocity

Fig. 5.3 Block Diagram Representation of the System of ODE's

The following Table 5.1 shows the individual subsystems defined in the above diagram.

(437.37 -

47.353 9)z2 (47.353 0 -

437.37)2i
Subsystem 1 Subsystem 2

-+GD
0ut1 Dutl

z2 z1

page 67
Chapter 5 Multi-degree of Freedom Model

(3.892 -

125.43 0) 2i(Q.853-szn0-
0.065 cos9)
-

Subsystem 4

Subsystem 3

zldotdot

cn ? 125^T^>
?) K""D +CJD
th 0ut1 Out1
Oain
2 0 .853'si n(u [1 ]>0 co<u [1 ])
th

02
0(0.1647 -

szn0
-

0.013 cos9) 4.57 -

84.45 0

Subsystem S Subsystem 6

d>

xr*<JD
Out1
3"
2 0 1 B47"si n(u [1 ]>D
. .0 1 cos(u [1 ])
th

(')'

0.1647 -cos0 + 0.013 -szn0 (277.07 22


-

951.55 21)

Subsystem 7
Subsystem 9

( 2 >- 0.1B47"oos(u[1])t-0.013,sinCu[1])
1 277.07
th *G-KI3
z2 Out1
(1 ) Gain

0ut1

(u[1]u[1])
2 851.55
thdot
z1

Gainl

(32312.48 0 -

1749.2 02) (106012.07 -

21
-

364076.59 z2
-

1109.31)
Subsystem 11

Subsystem 10

*Q-*CD
Out1

Table 5.1 Block Diagram Representation of the Subsystems

page 68
Chapter 5 Multi-degree ofFreedom Model

5.2.4 Response Analysis

A computer simulation of the system can be performed in Simulink to find the responses.
The following Fig. 5.4 shows the responses of
0, zu and z2 obtained by running the
simulation for 5 sees.

Response of Theta vs. Time Response of z1 vs. Time


0.2

0.15
-n

CO

CD 0.1
JZ

CD

C 0.05
Q.
to
CD
cr
0

-0.05

( 2 3 12 3 4
Time (sec) Time (sec)

Response of z2 vs. Time Superimposed Responses


0.015

0.01

0.005
CM
N

CO
0

S -0.005
E . _ _ UI - -
Aj - - J -^B l^ft - -
1 Ui - - Ui-
a.

-0.01

-0.015

2 3 12 3 4
Time (sec) Time (sec)

Fig. 5.4 Responses of 9, z\, and 22

It can be seen that the response of 0 has the highest amplitude oscillations (approximately
0.1662 rad = 9.52 deg), as expected from the model behavior. The response of 21 is

proceeding in the negative direction and is never positive because the center of gravity is
concentrated at a horizontal distance of 0.5 in from the center of the main wheel. The
lowest value of 21 is
-

0.0289 m. The response of z2 has a maximum of 0.011 m and

minimum of -

0.0114 m in the chosen time interval of the simulation.

page 69
Chapter 5 Multi-degree of Freedom Model

A phase plot representation of velocity vs. position of the three coordinates is shown in

Fig 5.5.

-0.015 0.015

Fig. 5.5 Phase Plots

For a specific set of initial conditions on 0, z\, and z2 (here the simulation is done by
zero initial conditions on 0, z\, and z2), the total energy of the system has a
assuming
specific value and the individual contributions from the three coordinates can be seen

from the above plots. It can be observed that the combination of the velocity and the

position to achieve that fixed amount of energy value changes between cycles. That is

why we are not seeing concentric loops of phase curves.

It is the three coordinates or the variation of one


interesting to see the interdependency of

coordinate with respect to the changes in the other two coordinates. This is depicted in

the following figure (Fig. 5.6).

page 70
Chapter 5 Multi-degree ofFreedom Model

Plot of Theta, z1 & z2

"*n
1

0.015 -s
*
-

fc
i

;
0.01 .

1
^-

-
^

jl J i- 7
. *
1

1
_
*
1
0.005 -V

'

E, i
y i I t
~

l\ A !
CM
-V'""i
/ i j \"**"i.\
Q-.
! -~

'n
A----"*"""

Ill -
* "
* . w !,
0.005 -
--
"

"*"-. *.V

-0.01

0.15
-0.02 0.1
0.05
-0.03
z1(m) Theta (rad)

Fig. 5.6 3-D Plot of 0, zu and 22

From the first graph, we can see that as 0 varies between 0 rad and 0.1662 rad , z\ varies

sinusoidally between 0m and 0.0289m. At the same time, 22 proceeds sinusoidally

with many lobes between the bounds 0.0114 m and 0.011 m. Another observation to

be noted is, a minor change in theta causes the response 22 react very quickly whereas 21

reacts slowly in comparison.

An FFT (please refer Appendix B.3 for FFT codes) is performed on the response signals

to analyze the frequency contents present. For this purpose, the Simulink model is run for
211
3.158 sees to generate N = = 2048 data points. The sampling frequency is chosen

as 670 Hz, which is more than twice the value of any inherent frequency present in the

three signals to avoid aliasing. The magnitudes of the Fourier Coefficients of the three

signals are plotted in Fig 5.7.

page 71
Chapter 5 Multi-degree ofFreedom Model

Frequency Contents in Theta, z1 & z2


20D

.150

100

50

0' o o I QfloodoodooQioooioooiooo
01234567B
Frequency (Hz)
30m 1 1 "T T T

.20

10

Ql

01
O O 0l 1I II A O QlQ O pip O O
OOOOdOQQ'QP
2345678
Frequency (Hz)
1
>
<j
_4

C
E2-

Q O <i? ? q> O Q r <7s 171 O O O i 0 0 01 0 p p POP


2 3 4
Frequency (Hz) ,9 Hz
'1.3 Hz

Fig 5.7 Results of FFT

The two dominant frequencies seen in all the three responses of 0, z\, and z2 are 1.3 Hz

and 4.9 Hz. This is quite predictable from the superimposed response plot of Fig. 5.4,
where we can see that all the three responses have the same period of 0.76 s. The second

dominant frequency, 4.9 Hz is more present in the responses of 21 and z2, which is that

of the small peaks seen in their responses. The FFT of z2 shows that there is a sinusoid of

higher (approx. 99.45 Hz) in its response, which is shown in Fig. 5.8.
frequency present

These higher frequency sinusoids represent the ripples or the disturbance signals present

page 72
Chapter 5 Multi-degree ofFreedom Model

in the response of z2 where as these sinusoids are not seen in the responses of 0, and z\.

FFT of z2 shows higher freqencies

CM

Dl 3
CO

Highbr Sinusoids
Frequency^

1 -

u \ . ~ --
-^~-l, *| M Ml | - - -' "

2D 40 60 BO
Frequency (Hz)

Fig. 5.8 FFT of 22 Shows Higher Frequencies

5.2.5 Analysis with Runway Profile

Now that we have analyzed the system in free vibration, i.e. without considering the

runway profile and the airplane velocity, we are now in a position to consider these

effects in to our analysis. Recall from Figure 5.1 that the coordinates of the points J and
H are (xq, zq) and (xh, zjf). As stated earlier, the assumed runway profile at the point J
is of the form:

20 = a-
sin(X xq) (5.20)

where xq is the displacement of the point J measured from the origin of the global

coordinate system and is given as:

xQ = v0-t

page 73
Chapter 5 Multi-degree ofFreedom Model

and

2-7T
A
I

Here a and I represent the amplitude and wavelength of the runway profile and vo is the
airplane velocity during landing.

Similarly, at the point H, the runway profile is given by:

zh = a sin(X xh) (5-21)

To use the above equation we need to derive an expression for xh in terms of the plane

velocity and the pitch angle. The x-coordinate of the point H from the origin of the

global coordinate system is given by:

xH =
xQ
-

QF sin9 + QP cos9 + PD sin9


-

ED sin(<f>
-

9) (5.22)

where:

sin (cf) -

9) = y/1
-

cos2(ct>
-

9) (5.23)

or:

fQFcoSe +
sin{
-6)^1-

QP^ne-Pn.cosSy ^

In the expression for sin(cj)


-

0), the effects of 2i and z2 are neglected. Therefore,


equation 5.21 becomes:

= o,
.

sm<
f2 -tt
-

x0 + (PD-QF)sin9 + QP-cos9 (5.25)


zh

(QF cos9 + QP sin9


-

PD cos9\
,

ED J

page 74
Chapter 5 Multi-degree ofFreedom Model

A Taylor series expansion of order 3 is carried out on zH to obtain a simplified form.


Accordingly the equations of motion now become:

Equation of Motion 1:

0 =
21 (0.853 sin9 0.065 cos9) + 21 (47.353 0 437.37) (5.26)
-
-

+ 22(437.37 -

47.353 0) + (3.892 -

125.43 0)
+ 0.65 cos9 -

8.42 sz'n0 szn(62.83 -

1)

Equation of Motion 2:

z\ =
0(0.1647 sin9 -

0.013 cos9) + (277.07 z2


-

951.55 21) (5.27)


02
0.1647 -cos9 + 0.013 sin9 + (4.57 -

84.45 0 -

6.94)
+ 11.54 sin(62.83 t)

Equation of Motion 3:

22 = (106012.07 21
-

364076.59 z2
-

1109.31) (5.28)
+ (32312.48 0 -

1749.2 02)
02
+ 645.16 -

1004.6 -
sin(62.83 t+ 1.714)
02
+ 1138.53-0+ 11419.55 cos(62.83-t+ 1.714)

By looking at these equations, we can see that they have the same basic forms of the

equations of motion derived without considering the runway profile. The addition of the

runway profile has only introduced the inputs and its derivatives into these equations.

The system is again simulated in Simulink and the modified responses are shown in the

following Fig. 5.9.

page 75
Chapter 5 Multi-degree ofFreedom Model

Response of Theta vs. Time Response of z1 vs. Time

2 4
Time (sec) Time (sec)
Response of z2 vs. Time Superimposed Responses

-0.015

2 4
Time (sec)

Fig. 5.9 Responses of the System with Runway Profile

From the graphs, we can see that the amplitudes have been slightly changed. There is not

much change observed in the maximum value of 9. But the minimum value of 9 is

changed from 0 rad to 0.0049 rad as a result of the runway profile. As before, z\ is

proceeding in the negative direction. The amplitude changes in the response of 22 is very

small of the order of


j^
of a m. The phase plot representations of the three

coordinates are shown in Fig. 5.10.

page 76
Chapter 5 Multi-degree of Freedom Model

2 0

0.0B 0.08 0.14 0.16 0.18


Theta
1 i i 1 1 i
: :

0.5
__
i^r
^jf

r|h{ ^tffTjj^^V+T13SS??^??!J Yzjky) Yjf)


--

S-=sgg2^gl^sSSSie?
-0.5

i i i i i i i i i
-0.04 -0.D35 -0.03 -0.025 -0.02 -0.015 -0.01 -0.005 0.005 0.01
z1

-0.015 0.015

Fig. 5.10 Phase Plot Representations of 0, z\, and 22

the system with and without the runway profile we


By superimposing the responses of

is The effect of the


graphically how the behavior of the system
can see changed. runway

profile on 0 is observed to be very small. But 21 and z2 have considerable changes in their

behavior. It can be seen that the modified responses have the same trends of their

counterparts without the runway profile. In other words, the new responses have the

in the case of the old ones. In the case of


same increasing and decreasing nature as seen

21 and 22, we can see that signals of higher


frequencies have been introduced, which

cause the responses to react with more peaks in a cycle as compared to that of the system

with no runway profile. These changes are clear from the following superimposed plots

(Fig. 5.11).

page 77
Chapter 5 Multi-degree of Freedom Model

Response without the Runway Profile


Response with the Runway Profile

Time (sec)

With Runwa^-Pj-ofile
0.015 -.
thg

0.01 -,
-"

0.005-

0-
IK.
-0.005 -
"\ With_dyLthe-Ru'riway"ProfiJd ~^vt-._
-0.01 -

-0.015^
0

-0.04 Q
z1(m) Theta (rad)

Fig. 5.11 Superimposed Responses

page 78
Chapter 5 Multi-degree of Freedom Model

As a final step of the analysis, let us perform the FFT on the modified signals. The
modified system is run for 6.024 sec in Simulink to generate N = 3750 data points. The

sampling rate is chosen as 600 #2. The reason for choosing this particular combination

is, to make sure that all the relevant frequencies are included in the sampled signal. The
FFT output for the first 75 points are shown in Fig. 5.12.

Frequency Contents in Theta, z1 & z2


400

300<>-

200 -

100

txtxsfff I'famrocoraraoaf^
10 12
Frequency (Hz)
B0

.40

20

OCQCtfffl >\<2xxhaxxx0XcaaT&?? 'foaraiMaoooorexxxxxito <XrX033XXXXX>


4 E 10 12
Frequency (Hz)
10

rN 6

qJcSffltf TftfoXaiCQ^^ iTWt^wtogorxcffjr^^ l(itotBP?rTCCOCCO


' 6 8 10 12
V \
approx. 1.20 Hz
\ Frequency (Hz)
approx. 4.64 Hz
approx. 9.6 Hz

Fig. 5.12 FFT Outputs plotted for First 75 Points

It can be seen that the cyclic frequency of 1.3 Hz without the runway profile has been

changed to 1.28 #2. This is clear from the graphs of the response comparison (5.11)
also, as the time periods of the modified responses have increased by a small amount.

The second dominant frequency of 4.9 #2 has changed to AM Hz although this

frequency is not clearly seen in the response of 0. The sinusoidal frequency of the runway
profile (i.e. w = 2 it 10 rad/'sec or / =
10 Hz) can also be seen in the cases of 21 and

page 79
Chapter 5 Multi-degree of Freedom Model

z2. As we have seen before, the response of 22 shows higher frequency sinusoids ( 93

Hz), which is graphed in the following Fig. 5.13. These higher frequency sinusoids

correspond to the ripples or the disturbance signals present in the response of 22.

FFT of z2 shows Higher Sinusoids


10 1 1 1 1 1 1 1 1 1

CO
E

3 -

[ ! 1 ] i Li....
t\X 10 20
!
30
.

40
1
1 _j_

50
1
BO
i
1

70
1
1

BO
!
1 i>L

90
hi
100

Frequency (Hz)

Fig. 5.13 FFT of z2 Shows Higher Frequencies

A comparison between the calculated FFT coefficients of both the systems can be

observed from the following superimposed plots (Fig. 5.14).

page 80
Chapter 5 Multi-degree of Freedom Model

Frequency Contents in Theta, z1 & z2


400 1 1 1
FFT without the Runway Profile
.300
-

1
FFT with the Runway Profile
, ,

i i i
200

100 Ii! i i i !
iZAj ! ; |
10 12
Frequency (Hz)
BO 1 i i

FFT without the Runway Profile


FFT with the Runway Profile
.40

E 20 I ; ; ;

La =: iv-w_ 1 i /v i
10 12
Frequency (Hz)
15 I i i

FFT without the Runway Profile


FFT with the Runway Profile
.10

2L
en
aj
E
5

.zJl
I
i^. j
"A! ._ i ^~-_

10 12

Frequency (Hz)

Fig. 5.14 Superimposed FFT Plots

As the by the FFT algorithm correspond to integer multiples


frequency values computed

kth
of
j4fj Hz,
the output from the FFT corresponds to the frequency of jj^Hz.

to decide the exact combination of the sampling


Therefore, it requires trial and error

frequency, , and the number of data points, N, for the FFT to show the peaks exactly on

the dominant frequencies present in the signal. But even with this point in mind, we can

see that the peaks of the system with the runway profile are shifting towards the left side

of the graph, meaning that there has been an increase in the time period of the response

signals.

page 81
Chapter 6

DAMPED SYSTEM

6.1 Introduction

The characteristic behavior of the response of a dynamic system relates to the stability of

the system. We saw that the response of the system without a damper is purely oscillatory
with constant amplitude because there were no dissipative forces considered in deriving
the equations of motion. For such a system, the poles of the characteristic equation lie on

the imaginary axis of the complex plane and this type of system is said to be neutrally

stable. In order to bring the system to a standstill the total energy of the system should be

converted to some other form of energy by dissipation. This is achieved by providing a

damper in the suspension assembly. Since the prototype model does not have a damper in

its nose gear assembly, a simple linear damper or dashpot is considered in the nose gear

assembly. For a linear damper, the damping force generated is directly proportional to the
difference in velocity of the two end points of the device.

6.2 Equations of Motion of the Damped System

6.2.1 Forces Due to Dissipation

According to Lagrange's approach (Chapter 2), the loss due to viscous friction devices is

represented by an energy dissipation function, D, which depends upon the velocities of

the system and the damping constants. Therefore, we have

*=/(*?)
coordinates.
where qi axe the generalized velocity

For optimum stability and performance, the damper is placed along with the spring as

shown in the following Fig. 6.1.

page 82
Chapter 6 Damped System

Fig. 6.1 Model with Damper

As shown earlier, the link AC is subjected to a displacement of As = BD DC coscj)

Ls. Substituting for coscj), we get:

DC
A< = BD ((QF -

PD) cos9 + QP sin9 + zx -

z2)
-

Ls (6.1)
ED

Since the damper is placed on the same link, the total velocity of the damper is given by
the time derivative of As. Therefore,

^^ = -

^ -

( -

(QF -

PD) sin9 -9 + QP- cos9 9 + zx -

z2) (6.2)
dt ED

and the dissipation function is given by:

1 (DC QP- cos9


(QF 9 + zx (6.3)
( PD) +
j
-

sin9
z2)
- -

-9

where c\ is the damping constant.

page 83
Chapter 6 Damped System

The forces due to dissipation in the three be found


coordinates can by taking the

derivatives of the dissipation function with respect to the particular coordinate. Thus we

have:

dD (DC\
Cl
'
(QF -

PD) sin9 -9 + QP- cos9 9 + zx -

z2 (6.4)
d'9 \EDJ
'
-

(QF -

PD) sin9 + QP cos9

(DC\2
dD
~Cl'

(QF -

PD) sin9 -9 + QP- cos9 0+ zi


-

*2 (6.5)
dzx \EDJ
(DC\2
dD
(QF -

PD) sin9 -9 + QP- cos9 9 + zx -

z2 (6.6)
dz2 \EDJ

6.2.2 Equations of Motion

The Lagrange's equations of motion for the damped system are given by:

dL dD
0
dt\d9J ~d9+~&)
d f dL dL dD
0
dt \dzi dz\ dz\
d f dL dL dD
0
dt \dz2 dz2 dz2

By evaluating the terms in the above set of equations and substituting for the data, we

arrive at the three equations of motion representing the damped system.

Equation of Motion 1:

0 =
21 (0.853 sin9
-

0.065 cos9) + 21 (47.353 0 -

437.37) (6.7)
+ 22(437.37
-

47.353 0) + (3.892 -

125.43 0)
9
cos2

+ ci 0(0.005 sin9 cos9


-

0.023 -

0.000275)
+ Cl(2l-22) 0.00834 sin9
-

0.077 cos9

+ 0.65 cos9
-

8.42 sin9 sin(62.83 t)

page 84
Chapter 6 Damped System

Equation of Motion 2:

21 =
0(0.1647 sin9
-

0.013 cos9) + (277.07 951.55 (6.8)


-

z2 zx)
(*)'

+ 0.1647 -cos9 + 0.013 sin9 + (4.57 02


84.45 0 6.94)
- -

+ ci 0(0.0016 n^ -

0.015 cos0) + 0.0488 ci(z2


-

*i)
4-
11.54 sm(62.83 t)

Equation of Motion 3:

22 =
(106012.07 21
-

364076.59 z2
-

1109.31) (6.9)
+ (32312.48 0 -

1749.2 02)
+ a 0(5.692 cos0 -

0.616 sin9) + 18.675 cx(zx


-

z2)
02
+ 645.16 -

1004.6 sin(62.83-t+ 1.714)


02
+ 1138.53-0+ 11419.55 cos(62.83-t + 1.714)

The above system of equations is simulated in Simulink using damping a constant,

ci =
20 N -

s/m and the following responses are obtained (Fig. 6.2).

page 85
Chapter 6 Damped System

40
Time (sec)
0.01

% 0.005 |
CN

8 o
c
o

-0.005

cr

-0.01

0 10 20 30 40 50 BO 70 ao
Time (sec)

Fig. 6.2 Responses of the Damped System

The nature of these responses shows that the system is underdamped and it comes to a

stable state of oscillations after a reasonably large amount of time. Even if we


steady
to introduce damping to the system, it can be seen that the
increase the value of V more

the steady decreases but at the cost of a powerful damper which


time to reach oscillations

can dissipate the energy more efficiently. Another alternative to provide more damping to
the system is through a controller.

A comparison of the responses of the undamped and damped systems (Fig. 6.3) is useful

to understand the effect of damping on the system.

page 86
Chapter 6 Damped System

Undamped System
Damped System

'

'MflT:1:ir

i tu
_L

10 20 30 40 50 60 70 80
Time (sec)

Fig. 6.3 Damped vs. Undamped System

6.3 Simulation of an Actual Landing Gear

6.3.1 Landing Scenario

The landing of an aircraft begins with an initial approach, [2] (see Fig. 6.4). Flaps are

used for most landings to permit a lower -


approach speed and a steeper angle of descent.
The airspeed and rate of descent are stabilized, and the airplane is aligned with the

runway centerline as the final approach is begun [11]. When the airplane descends across

the threshold of the runway, power is reduced further and the rate of descent and airspeed

are slowed down by the pilot. The airplane is kept aligned with the center of the runway

mainly by use of the rudder.

page 87
Chapter 6 Damped System

=^=$

Fig. 6.4 Stages of Landing

The pilot's objective is to keep the airplane safely flying just a few inches above the

runway's surface until it loses flying speed. In this condition, the airplane's main wheels

"squeak
on"

will either or strike the runway with a gentle bump. With the wheels of the

main
landing gear firmly on the runway, the pilot applies more and more back pressure on
the control wheel. This holds the airplane in a nose-high attitude which keeps the nose

wheel from touching the runway until forward speed is much slower. The purpose here is
to avoid overstressing and damaging the nose gear when the nose wheel touches down on

the runway. The landing is a transition from flying to taxiing.

6.3.2 Landing Simulation

As far as the landing simulation is concerned, there are two cases of interest: (1) the

simulation which starts when the main wheels just touch the runway, and continues until

the nose wheel touches the runway (Fig. 6.5), (2) the simulation which starts when the

nose wheel touches the runway, and continues until the aircraft comes to a standstill.

Consider the first case, when the main wheels just contact the runway with the nose

wheel being still airborne. To actively dissipate the dynamics during the touchdown

impact, a linear damper is considered with the main wheel suspension spring. Since the

nose wheel is airborne, the displacement, 22, is zero and the pitch angle, 0, is a variable.

During an actual landing touchdown phase, a lift force at the tail wing is created by
controlling the flaps on the tail wing [12], [13]. This creates a nose down pitching

moment. The consideration of the pitching moment in the analysis is very essential during
the landing touchdown phase because the nose of the airplane is airborne. Since the center

of gravity of the plane is at the aft position, the resultant upward force will lift the nose of

the plane upwards. However for the analysis purpose, we will consider the effect of the

pitching moment in terms of a variable (say Pm) in the equations of motion and a suitable

value will be determined when we discuss the simulation of the overall landing scenario.

page 88
Chapter 6 Damped System

Another variable acting is the displacement of the center the


of main wheel, 21 The total
.

potential energy consists of the potential energy due to


gravity and the potential energy
due to the main wheel spring. The height of the center of the nose wheel from the static

equilibrium position can be found as:

hnc =
zi + (QF -

PD)cos9 + QP ED 9)
sin9
(6.10)
-

cos(cj) -

Fig. 6.5 Landing With Nose Wheel Airborne

Therefore, we have:

P.E =
(M(z0 + Z-L + QF- cos9 + QG sin9) + Mm-z1)-g (6.11)
+ Mn(zx + (QF -

PD)cos9 + QP sin9
-

ED cos(cj)
-

9)) g
z0)2

+ -

km(zi -

Similarly, the kinetic energy is given by:

page 89
Chapter 6 Damped System

KE= -M (vq-9-QF- cos9 9QG-


(6.12)
-

+ {z0 + zi-9-QF-
sin9 + 9-QG-

\lG{0)2 \Mm(zi)2
+ +

The Lagrangian becomes:

L KE-
=
PE
(6.13)
L=-M (vq-9-QF- cos9 -
9-QG-
2

+ {z0 + zx-9-QF-
sin9 + 9 QG

\lG(?)2 \Mm{zl)2
+ +

M(z0 + zi + QF cos9 + QG sin9) +


Mm-
zx
-

Mn(zx + (QF -

PD)cos9 + QP sin9 -

ED cos(cj> -

9)) g
Zq)2

X km{zX -

Additionally, the dissipation function is given by:

D = -

C2(ZX
-

Zq) (6.14)

where c2 is the damping constant of the main wheel suspension damper.

Therefore, the dissipation force in 21 coordinate is:

dD
=
C2(zX
-

Zq) (6.15)
dzx

Since the coordinate 0 is measured positive in the anti-clockwise direction, the nose down
pitching moment Pm is negative. The Lagrangian represents a 2-dof system in 0, and 21

coordinates. Therefore, the two equations of motion are given by:

page 90
Chapter 6 Damped System

d (dL\ dL dD
=

dt{ti)--dl+ti -Pm{t) <6"16)


d_(dL\
dL_ dD_
dt\dzxJ dzx dzx

which, after evaluation becomes:

Equation of Motion 1:

0 (zq + 0.065 cos9) + 8.372


=
2i)(0.853 -
sin9 -

sin9 -

0.684 cos9 (6.17)


-

0.006 sin(
-

0.3397 + 0) -

0.4366 Pm(t)

Equation of Motion 2:

(0)2

2j =
0(0.1647 sin9
-

0.013 cos9) + 0.1647 cos9 + 0.013 -


sin9\
(6.18)
-

0.0843 km(zx -

zq)
-

0.0843 c2(2i
-

z0)
-

0.998 z0
-

9.835

The angle cj> is assumed to be constant from the time instant when the main wheels

contact the runway until the nose wheel touches the runway. The value of 4> in the

nosewheel airborne position can be found from the expression for the static deflection of

the spring ks given by:

AS =
BD-DC coscj)
-

Ls (8.19)

Since the weight of the nose wheel is causing the spring to be unstretched, we have:

As =
0 = BD -

DC coscj) -

Ls
,
BD -

Ls
cos0=___
or:

19.46
or: cj) = = 0.3397 rad

The above equations are simulated for 1 sec, assuming the damping constant for the main

wheel damper, c2 =
5N s/m, and the responses are given in Fig. 6.6. The value of Pm
is determined in such a way that the pitch angle 0 becomes zero after 1 sec of simulation.

The corresponding value of Pm is found to be 0.31 iV m.

page 91
Chapter 6 Damped System

Responses at Touchdown
0.15

D D.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1


Time (sec)
0.01

0
f V '
/ \ ' i / i i i /
/i '
/ '
\ ' '
/ \ '
'/

-0.01 t t^t * r y---T---\--T -i- 1- J---/


w
(!)
C
o -0.02
Q.
CO
QJ
-0.03
IV--A--J ' ^t_ .i
A.-/ i i l
cr

-0.04

0 0.1 0.2 0.3 0.4 D.5 0.6 0.7 0.8 0.9 1


Time (sec)

Fig. 6.6 Responses at Touchdown

The 2-DOF system is simulated until the aircraft loses its forward speed. At this point, the

nose wheel touches the runway, which introduces the coordinate 22 into the model,

thereby making it a 3-DOF system. In Simulink, to simulate the system continuously

considering the two cases, we need to make use of the enable feature. If we place the

enable block in a subsystem, it means that the subsystem operates only when the control

signal is positive. In the first part of the simulation, i.e. when the main wheels are on the

runway, a positive control signal is fed to the 2-DOF system, making the system enabled.

Similarly in the later part, i.e. when the nose wheel touches the runway, a positive control

signal is fed to the 3-DOF system, making the system enabled. The two outputs are then

merged using a merge block. The complete landing simulation is shown in Fig. 6.7.

page 92
Chapter 6 Damped System

th thdot r) zidot
A m
DjU Store Djtj Store DjLi Store Pita Store
Memory Memoryl Memory3 Memory2

|* 0JQ 51371
PUnt Model Read2

Read3

Logic Box
In Data Store
Readl
&

Data Store
r Wrjl2
WUIfZ.
j
I | -.^L-L^u.dot

KE of the Plane

Oata StSi
WrfteS

P.E of the Plane

Spring Force

2-DOF Subsystem Model (Kd/vO)

p^m
j

* tm<i

-CZD

thdotdol ztdotdot 4 1 dotdot


<^i z1 dotdot]>^
>p~*
>fl><

ODE of Thdotdat

ODE of zldotdot

page 93
Chapter 6 Damped System

3-DOF Subsystem Model (t>=d/vO) -KID


a

K3
-U|>KI3
LiMd^>-
z2
Integration of zldotdt
Integnbon of Thdotdi I Integration of 22dofafi 1 z2dot
*^
riant

->^ld*1

ODEcfztdotdot DDE ofifidotdol


ODEofThdotdol

Integration ofzldotdot
Integration of Thdotdot
r+CD
Zldot

rH^ | kjc] (T>


I Goto I ooid zldotdol

Q>
Integrator thdot
-KID ridoi.
^ zlJcV n

Fiom
CZD*
thdot

Logic Box

Integration ofzldotdot

<3>
time -KID
time>=d/i>0
Relational
Operator!
1 ? -KD
z2dotdot z2

-KTD

Relational
Operator

page 94
Chapter 6 Damped System

Thdotdot Model

-CZD

zl

zldotdot

-dD

th .

thdot i

thdot

z2dot
<U
ildot I i z2dot

Lqd zldot

zldotdot Model

page 95
Chapter 6 Damped System

z2dotdot Model

*n 4
zh

th <-

thdot *f-

thdot

zh Model

rqrt(1-^2.5g"rp.033-co<u[1J)f0.30'sin(u[1]>ur2]-uP]y2)WJ-rMS^s

th
Z1

z2

Fig. 6.7 Landing Simulation

The purpose of the logic box (shown in yellow color) is to divide the time frame into two
segments. Basically, it compares the simulation time with the time taken for the aircraft to

(
travel a specific distance on the runway ;
where d is the distance traveled, and v q is the

landing speed) before the nose wheel lands on the runway. The output of the logic box is

used as the control signal to control the two subsystems representing the two cases of the

landing scenario. The orange color block represents the 2-DOF model, which runs in the

initial part of the simulation and the cyan color block represents the 3-DOF model, which

page 96
Chapter 6 Damped System

runs in the later part of the simulation. The outputs of the simulation are shown in Fig.
6.8.

ResponsBs During Touchdown and Taxiing

0.01

.-0.01 ___-.J_-_-.l
-
l________--_l __._l.__ A _ - _ J _ ........ --------_.

-0.D2

10 20 30 40 50 B0 70 00
Time (sec)

Fig. 6.8 Landing Responses

It is to be noted that the above simulation is run with the initial condition on

9 =
hdeg = 0.0873 rad. The initial conditions for 9, 9, zx, and zx in the 3-DOF model

are inherited from the final values of the responses of 9, 9, zx, and zx in the 2-DOF model

(Fig. 6.7). This is achieved by defining state ports in the integrator blocks of the 2-DOF

model and these signals are fed to the integrator blocks of 9, 9, zx, and zx in the 3-DOF

model. Another feature used in the simulation is the memory block. Since Matlab does

not allow the output of the merge block to be used for feedback, the partcular signal is

page 97
Chapter 6 Damped System

is stored in a temporary memory block and then accessed


using read from memory
feature.

6.3.3 Energy Transition

It is helpful to study the behavior of the total energy of the system under the whole

landing scenario. The total energy, which is the sum of the potential energy and the

kinetic energy, has smaller amplitude in the initial stage of the landing as compared to the

later stage. This is because in the initial stage, there are only two variables (9, and zx)

contributing to the total energy. Since the main wheel damper is acting in the initial

stage, the total energy is dissipating, which is quite predictable from the nature of the

response of zx . As the nose wheel lands on the runway, there is a sudden increase in the

amplitude of oscillations of the total energy due to contributions from 9, and z2

coordinates. At the same time the damper starts dissipating the energy, thereby
converting the total energy of the system into other forms. So we can see that the

amplitude of total energy is continuously decreasing until it reaches the steady


oscillations. The energy models and the responses are shown in Figs. 6.9 and 6.10.

KE of the Plane

KE0>=dAr0)

page 98
Chapter 6 Damped System

KEt<dNO LLU
Enable

QF"oo<u[1])fQS"5in(u[1])
X 1
Fen
Scope
th
thdot

ii]-2+ur2i*2
_J^>-Jm> KCV-U(^>rn
*^r
'
.QF"sln(ii[1]>-aG"cos(ij[1D 7~Z. *

Gain4 t"T,
Gain3
? KE(t>=oA
E(t>=dA0)
X ?

G>
zOdtrt

Q>
zldot

unruin *K> W^> ?

GainB oainS

"[ir>i[i] >fv>--*'
*R>
Gain2 Gainl

KE1>=d/vO
E

Constant
onstant L|> +

QF*co^u[1]>-QG'siri(u[1D ?
Scope
th
thdot
f> j[1J^2+u[2)2

-4?H^*9
4J>>CD
-QF"ani:u[1I>aG-C0!<u[1D
KEQMdMQ
Gain4 Oain3

cz>
zOdot

Q>
zldot

U[1]u[1] ?"'
H^
GainO GainS

u[1]"u[1] *f2> W"^> ?

Gain2 Gainl

CD- ?> ?
u[1]"u[1] H^>
z2dot
GainIO Gainfi L-1

page 99
Chapter 6 Damped System

PE of the Plane

PEfJ>=dArO)

PEKd/vO m
.n

Q> (QF.PD)roos(u[1])i-QP"sir<u[1I>ED"o<M(u[1l.phi)o KC) ?K>+*


Scope

X ftaim

z1
f>M+Mm -*U>->9-

J>>-KZD
*r pe (t<dMi:
.

QP(u[1l)*Q-sin(u[1D
?) >
?

Font T
C>
zO

inQ ? u[Tu(1] *R> H^


T Fcn6 Gain3 Gain4

page 100
Chapter 6 Damped System

PEt>=d/vO
0
?

GD-* BCtOF.P07co<u[1I>aP"jin(u[1Du[2>uprxED
-{*>^[>-K5)-M^>KD
^
zl T FcnS
Fcn5
iT.
Gain12 Gainll
X PE(t>"07>
Ert>=o'.0)

z2
BO-Ls I

*>-> KO
|"~
Gain7

QPco<u[1I)^QG'sir<u[1D 9~*E>
Fcr>4

^ -+) ? "iTii] *f> *f>> *0


''
J FcnB Gain3 Gain4

KO ? "[Ifli'll *t3> H^5"


T Fcn7 Gain5 GainO

Fig. 6.9 Energy Models

Separate models are created to represent the total energy in each case of the simulation

and are run conditionally using the enable feature. The outputs are then merged using
merge blocks.

The average total energy of the system in the whole landing scenario can be found by
taking the mean of all the energy values. The mean energy in the initial stage of landing is
found to be 611.3 Joules and it settles down to a lower value of 610.5 Joules towards

the later part of the simulation. This shift is due to the effect of damping on the system. In

other words, as the damping forces are negative due to their inherent dissipative nature,

they carry away some part of the total energy for dissipation. If we take a look at the

potential and kinetic energies of the system, we can see that eventhough the potential

energy of the system is not much affected by the introduction of the coordinate z2, the

kinetic energy is very much influenced by the contribution from z2.

page 101
Chapter 6 Damped System

Energy Variation in Landing


21

u>
20 j___ ___ *__. ------
-.---.__-__j__.

J, 19
ui
a- 1B

17
10 20 30 40 50 60 70 BO

B10

630 I 1 1
j I

620

610

LU
600

i i 1 ! ! i
590
10 20 3D 40 50 60 70 BO
Time (sec)

Fig. 6.10 Total Energy of the System During Landing

page 102
Chapter 7

CONTROLLER DESIGN STRATEGY

7.1 Introduction

The significance of control systems on modern machines/equipment can never be


underestimated because of their capability to influence a system to achieve some desired
performance. We have seen that the response behaviors of the multi-degree of freedom
aircraft landing gear model are
oscillatory and require a considerable amount of time to

settle down. We could imagine, how uncomfortable it would be if this system were used

in a suspension model by itself. In general, a dynamic system is used with a controller in


order to achieve desired output requirements. The objective of designing a controller for
the developed aircraft supension model is to ensure smooth landing in a normal landing
situation.

Control systems are generally classified according to the information used to compute the

controlling action. If the controller does not use a measure of the system output being
controlled in computing the controller action, the system is called open-loop control. If

the controlled output signal is measured and fed back for use in the control computation,

the system is called closed-loop or feedback control [14]. In practice, most of the

dynamic control systems are feedback control systems due to their inherent input tracking

nature. A simple example of an elementary feedback control system is shown in Fig. 7.1.

Disturbance
Plant

Control
Reference Signal Output
Input
->- Controller Actuator Process
Filter

Sensor

Sensor
Noise

Fig. 7.1 Block Diagram of a Feedback Control System

page 103
Chapter 7 Controller Design Strategy

The central component of a feedback control system is the process, which is designed for

a specific purpose and whose output needs to be controlled. The disturbance to the

process is the extraneous input which enters the process. The term actuator refers to the

device that can influence the controlled variable of the process. The combination of

process and actuator is called the plant, and the component that actually computes the

desired control signal is called the controller.

7.2 Control System Design

7.2.1 Steps in the Design Process

The design of control systems is a specific field of engineering design. The goal is to

obtain the configuration, specifications, and identification of the key parameters of a

proposed system to meet an actual need. The control system design process is

summarized in the flow chart [15] shown in Fig. 7.2.

Establish control goals

Identify the variables to control

Write the specifications

for the variables

Establish the system configuration

and identify the actuator

Obtain a model of the process, the

actuator, and the sensor

Describe a controller and select

key parameters to be adjusted

Optimize the parameters and

analyze the performance

If the performance does not If the performance meets

meet the specifications, then the specifications, then

iterate the configuration and finalize the design.


the actuator.

Fig 7.2 The Control System Design Process

page 104
Chapter 7 Controller Design Strategy

For the aircraft landing gear model under consideration, the goal is to control the response

behaviors to ensure smooth landing in a normal


landing situation. Therefore, our control

variables are 9, zx, and z2.

7.2.2 Transfer Functions

The knowledge of transfer functions is very essential before we detail the controller

design procedure. The transfer function of a linear system is defined as the ratio of the

Laplace transform of the output variable to the Laplace transform of the input variable,

with all initial conditions assumed to be zero. The transfer function of a system represents

the relationship describing the dynamics of the system under consideration. Generally, the

block diagram representations of linear dynamic systems use Laplace variables to

symbolize the different signals. We can represent the open loop and closed loop systems

in the block diagram forms as shown in Fig. 7.3.

R(s) +r^ Y(s)


-J
Eh>.
*
C(s) G(sj

R(s) Y(s)
C(s) G(s) H(s)

Fig. 7.3 Block Diagram Representations of Open Loop and Closed Loop Systems

is by:
For an open loop system, the transfer function of the system given

T^ =
W)=c{syG(s) (7.1)

be derived follows:
Similarly the transfer function of a closed loop system can as

E(s) =
R(s) -

H(s) Y(s) (7.2)

Y(s) =
G(s) C(s) E(s) (7.3)

Substituting for E(s) in Eqn. (7.2) from Eqn. (7.3) and rearranging, we get:

page 105
Chapter 7 Controller Design Strategy

TM -
y(s) C<> G^
(74)
-

V ;
'

l + [ }
R(s) H(s)-C(s)-G(s)

Eqns. (7.1 and 7.4) represent the loop transfer functions for the
open and closed
loop
block diagrams shown in Fig. 7.3. When the system equations are simultaneous
ordinary
differential equations, the transfer function that results will be a ratio of polynomials. If
we set the denominator of Eqn. 7.4 equal to zero, then the resulting equation is called the

characteristic equation of the dynamic system. The corresponding values of s in the


complex plane, which render the denominator equal to zero, represent points where T(s)
is infinity. These s -

values are called the poles of T(s). Values of s which will make the

numerator of Eqn. 7.4 zero are called zeros of T(s). These poles and zeros, play very
important roles in determining the characteristic responses of a system. Since the Laplace
transform of an impulse function is unity, the impulse response of a system is given by
the time function corresponding to the transfer function of the system. So we call the

impulse response the natural response of the system. The poles and zeros can be used to

compute the corresponding time response and thus identify time histories with pole

locations in the s -

plane. Consider a system, which has the transfer function,

H(s) = -

(7.5)
s + cr

and the impulse response is given by the exponential function:

h(t) = e^ty). (7.6)

When a > 0, the pole is located in the left half of the complex plane (i.e. s < 0), the

exponential expression decays and we say the impulse response is stable. If o < 0, the

pole is in the right half of the complex plane (i.e. s > 0). Because the exponential

expression here grows with time, the impulse response is referred to as unstable. The

time constant of a first order system is defined as the time when the response is \ times
the initial value and is given by:

r=-. (7.7)
(7

page 106
Chapter 7 Controller Design Strategy

Hence it is a measure of the rate of


decay of the response. To summarize, the coefficient

oft, in the exponential term of the response (o here), determines the rate of decay of the
response. The larger the o, the faster the decay and vice versa. In other words, the pole

locations corresponding to larger absolute values of s in the left half of the complex plane

represent a faster decay and a stable system.

7.2.3 Time Domain Specifications

Specifications for a control system design often involve some requirements associated

with the time response of the system. The requirements for a step response are expressed

in terms of the following standard definitions illustrated in Fig. 7.4.

*P i%

*H
i

09
'

0.1
't
I ^ ft

Fig. 7.4 Definition of Rise Time, Settling Time, and Overshoot

The rise time tr is a measure of the swiftness of the response. For a second order system,

the rise time from y = 0.1 to 0.9 is approximately untr = 1.8. Thus we can say that

1.8
(7.8)
un

The above empirical form is only accurate for a second-order system with no transfer

function zeros; for all other systems it is only a rough approximation to the relationship

between the rise time, tT and the natural frequency, un.

The settling time ts is the time it takes the system transients to decay. This is the time

required for the response, y(t) to reach the steady state. If we take a 1 % steady state

page 107
Chapter 7 Controller Design Strategy

envelope, we can define the settling time as the value of ts the


when decaying exponential
reaches 1 %. Thus we have:

4.6 4.6
ts .

(7.9)

where a is the negative real part of the pole.

The overshoot
Mp is the maximum amount the system overshoots its final value divided
by its final value (often expressed as a percentage). Analytically, overshoot occurs when

the derivative of the response is zero and from calculus we can arrive at the formula for

Mp as:

Mp =
eV1-<2! 0 < C < 1, (7.10)

and the peak time tp, which is the time to reach the peak is given by:

7T
Zp (7.11)
Ud

where Ud =
un y/l C,2, is the damped natural frequency.

It is to be kept in mind that these design requirements are only true for a second order

linear system; for more complicated real time nonlinear systems we can only use these as

guidelines in designing the controller. The graphical representations of the above criteria

can be delineated as shown in Fig. 7.5.

sin-1
1
klm(s) Um(s)

~T~

~R?ti ~R~fls>

(c) I (d)

Fig. 7.5 Regions in the s -


plane delineated by the transient requirements: (a) rise time; (b)
overshoot; (c) settling time; and (d) composite of all three requirements

page 108
Chapter 7 Controller Design Strategy

1.2 A Effects of Pole-Zero Patterns on the Dynamic Response

The basic idea of designing a controller is to influence the plant performance by placing
additional poles and zeros in the complex plane representing the root locations of the

system. By carefully manipulating these pole-zero locations, it is possible to influence the


system to achieve the desired performance requirements. We have already seen how poles

influence the dynamic response of a system. The effect of poles and zeros on dynamic

response can be summarized as follows [14]:

1 . Poles in the left half of the complex plane will cause the response to decay (or stable

response) whereas poles in the right half of the complex plane will cause unstable

response.

2. A zero in the left half of the complex plane will increase the overshoot if the zero is

within a factor of 4 of the real part of the complex poles.

3. A zero in the right half plane will depress the overshoot and may cause the step

response to start out in the wrong direction.

4. An additional pole in the left half plane will increase the rise time significantly if the

extra pole is within a factor of 4 of the real part of the complex poles.

7.2.5 A Perspective on Root Locus

The root locus method is a graphical representation of the migration of the roots of a

characteristic equation in the s -


plane. The root locus technique depicts how changes in

the characteristic equation and thus


one of a system's parameters will modify the roots of

modify the system's dynamic response.

Consider the closed-loop feedback control system shown in Fig. 7.3. The closed loop
is below:
transfer function of the system is given by Eqn. (7.4), which reproduced

,
,
Y(s) _
C(s) G(s)
1 [S) ~

R(s) 1 + H(s) C(s) G(s)

The characteristic equation, whose roots are the poles of this transfer function, is:

1 + H(s)-C(s)-G(s) = 0 (7.12)

page 109
Chapter 7 Controller Design Strategy

To come up with a form of the above equation suitable for analysis of the
roots, we put
the equation in polynomial form and select a parameter of
interest, say K. Assuming that
we can define component polynomials
a(s) and
b(s) so that the characteristic polynomial

is in the form a(s) + Kb(s), we can represent the transfer function by, ^. Now
L(s) =

the characteristic equation can be written as:

1 + K-L(s) 0
=
(7.13)

The above equation can be used for plotting the locus of all possible roots as K varies
from zero to infinity and use the plot to find out the best value of K. Furthermore, by
adding additional poles and zeros on this graph, we can study the consequences of

additional dynamics added to C(s) as compensation in the loop. The graph of all possible

roots relative to the parameter K is called the root locus. Some useful tips on root locus
are given below [16]:

1 . There is a single-valued branch of the root locus for each root of the characteristic

equation and the total number of branches is equal to either the number of poles or the

number of zeros, whichever is larger.

2. Each branch of the root locus starts at a pole, where K =


0, and ends at a zero, where

K =
+ oo. If the number of poles exceeds the number of zeros, there will be zeros at

infinity, equal in number to the excess. Excess zeros similarly means poles at infinity.

3. Along the real axis the locus includes all points to the left of an odd number of real

poles and zeros: no distinction is made between poles and zeros, and complex poles

and zeros are neglected.

4. If the number of poles


np exceeds the number of zeros nz, then as K approaches

infinity, (np nz) branches will become asymptotic to straight lines intersecting the

real axis at angles given by:

ft=(2*i)180!_ np-
nz

lfnz np, then as K approaches zero, (nz np) branches behave as above and
-

exceeds

originate at infinity along asymptotes.

page 110
Chapter 7 Controller Design Strategy

5. Branches of the root locus are symmetrical with respect to the real axis since all

complex roots appear in conjugate pairs.

6. Addition of a zero tends to attract the locus into the left half of the complex plane.

7. An additional pole moving in from the far left tends to push the locus branches to the

right as it approaches a given locus.

The review of control system theory provided here is very useful in analyzing the
response behavior of a dynamic system. As we have already derived the equations of

motion describing the dynamics of the aircraft landing gear system, we are now in a

position to design a controller to ensure smooth landing. Since the model under

consideration is nonlinear and complicated, most of the above described theory can be
used as only guidelines. In order to apply this theory to nonlinear systems, the method

followed here is to apply it to a simplified linear model of the same system, design a

controller for this linear model, and then make suitable modifications to the controller to

suit the needs of the nonlinear system. As an example to demonstrate this procedure, we

will consider the 1-DOF approximation (equation of motion in 9 coordinate) derived in

Chapter 4 as our nonlinear model.

7.3 Controller Design for 1-DOF Model

The single equation of motion (Eqn. 4.26) in the 9 coordinate derived in Chapter 4, is

modified into the following form if we add a linear damper along with the spring in the

nose wheel suspension:

9 = 20(0.853 sin9
-

0.0656 cos9) + 3.415


-

90.0234 -
9 (7.14)
-

cx 9 (0.0188 + 0.077 9 + 0.079 92)

Here z0 A sin^j^
is the runway profile. Eqn. (7.14) is simulated in Simulink
=
t)
and the following response is obtained.

page 111
Chapter 7 Controller Design Strategy

Response of the Damped 1 DOF Nonlinear Model


0.08

0.07

0.06

0.05

.0.04

\-\
lll::r:!::t:]:t:i::
0.03

D.02

0.01

8 10 12 14 16 18 20
time

Fig. 7.6 Response of the Damped 1-DOF Nonlinear Model

To linearize the above nonlinear model, we can make use of the LTI (Linear Time

Invariant) analysis feature of Simulink. The LTI model is created by defining input-output
ports in the model. This is shown in the following Fig. 7.7.

1-DOF Model with Damper


Q- timel

? Clock
To Woitepace2

Scopel

fz
Sine Wave
Scope

In1 th -+- ? th1

Input Point
Plant Model ^""'f.Watap

Product

D n(u [1 ]>0
.853"si co<u [1 ])

Fen

page 112
Chapter 7 Controller Design Strategy

Plant Model

Fcn1
Product!

Fig. 7.7 Linearization of the 1-DOF Nonlinear Model

The LTI analysis linearizes only that portion of the model which is contained inside the
input and output ports created. So it does not consider the type of input to the system.

After creating the input-output ports, a default step response of the linearized model can

be obtained by clicking on the Get Linearized Model button on the toolbar [17]. The

response of the linearized model is plotted against the original nonlinear response in the

following Fig. 7.8.

page 113
Chapter 7 Controller Design Strategy

Step Response
From. Input Point
0.08

Response- the-
of WtrnHnear-'r-ttof Mode)

yvywvwu

bTTthe"

Response Uneanzer] T-ddf Model

_L

15 25 3J

Time (sec]

Fig. 7.8 Nonlinear vs. Linearized Model

From the superimposed responses, we can see that linearized model is just a crude

approximation of the original nonlinear model. The steady state value of the nonlinear

response is approximately 3.4 times the steady state value of the linear response. Other

characteristics of the linearized model can be studied from the set of graphs shown in Fig.

7.9.

page 114
Chapter 7 Controller Design Strategy

Step Response Impulse Response


From: Input Point
From: Input Por6
0.025 015

0.i32 System modelj dof_wlth_z0_dBmper_2


Peak amplitude 0.0215
Overshoot (%): 935
E time-
P 0.015 At 0.337
0.05

0.0 1 y- "

System: modelj dot _with_z0_damper_2

DC gerr. 0.01 U

DISK -0.05

10 1J 20

Tme (sec)

Bode Diagram
From: Input Port Pole-Zero Map
10

> System: modelj dof_wlh_z0_demper_2 System: modelj dof_w8h_z0_d<


S - Pole: 88 + 9.43
-0.1

,;
Peak gain (dB): -11
Damping 0 0198
tf At frequency (rad/secj 9.48
Overshoot (%): 94
Frequency (rad/sec): 9.49

4
1 ^- '.* r "
?I
1 TTflT O
ns; : e

System: modelj dof_wlh_z0_d(


Pole: -0.1 B8-9.49i
3- -90

Damping 0 019B
Overshoot (%): 94
(rad/sec)-

Frequency 9.49

-0.5 0 0.5
10 10
Real Axis
Frequency (rad/sec)

Root Locus
I
i

Sys
< iam. 0
i
Pole: -0.1 BB 4-9.491

-i
Overshoot C%): 94
Frequency (rad/sec): 9.49

System: d
Gem 0
Pole: -0.1 BE
Damping: 0 D19B
Overshoot (%):94

i i _l
i
-1 D

Real Axis

Fig. 7.9 Characteristics of the Linearized Model

page 115
Chapter 7 Controller Design Strategy

The pole-zero map of the linearized model shows that there is a complex conjugate pole

at 0.188 9.49 i, which causes the system to overshoot with 94 %. At the same time

the settling time is 20.6 sec and DC gain is 0.0111. The asymptotes generated from the

two poles at 0.188 9.49 i seek zeros at infinity as per the root locus plot. Since we

know the poles of the system, we can create the open loop transfer function representing

the linear model as:

s -

( -

0.188 + 9.49 i)
s-(-
0.188 -

9.49 i)

which after simplification becomes:

T(s)
y)
= M. = l- (7.16)
R(s)
s2
+ 0.376 s + 90.0954

The characteristic equation of the open loop system is:

s2

+ 0.376 s + 90.0954 =
0 (7.17)

above equation with the generic form of a second order system


Comparing the

s2

+ 2{uns + u2n =
0, (7.18)

we can find out the DC gain, C, and un of the system. Thus we have:

= =
-111;
DC gain = =
&(,' + 0.376-
lim(^) s + 90.0954.) 9O0954

0.376 C 0.0198, and = 90.0954 =? un = 9.49 rad/ sec. Therefore,


2Cwn = => =
u2n

Cu)n = 0.188. To influence the system performance we need to modify the damping ratio

Since we have no freedom to alter these values for the


C, and the natural frequency un.

to study the effects of adding poles and zeros in the complex plane of
system we will try
the system.

page 116
Chapter 7 Controller Design Strategy

Let us say that we set the design criteria which require the response to have 25 %

overshoot and 1 % settling time of not more than 2 sec. To put these two conditions into
mathematical representations we proceed as follows:

For Mp < 25 % we have:

0.25 >e^~i2, (7.19)

which after simplification yields, C > 0.4037. To achieve ts =


2 sec or lesser for 1 %

settling time, C,un >


^ > 2.3. This implies that un > 5.697 rad/ sec. To design

controllers for the specific requirements, we will make use of the sisotool feature of

Matlab. The following three controllers are found to be successful to meet the specified

design criteria.

1. PD (Proportional Derivative) Controller

Since we know that the addition of a zero tends to pull the locus into the left half of the

complex plane, we will add a zero in the left half of the complex plane. The effect of

adding this zero is immediately reflected in the root locus in such a way that one of the

complex conjugate poles seeks the newly added zero and the second pole seeks the zero

at infinity. The Matlab sisotool is very handy in designing a reasonable combination to

achieve the desired performance [18]. The following values are obtained after trial and

error.

C(s) = 647(1 + 0.0833s) = 53.8951(s + 12),

which is of the standard form of a proportional derivative controller kp + kr)S. Therefore

we have, kP =
647, and kD = 53.8951. The closed loop transfer function model, the

outputs of the sisotool, and the response characteristics are shown in Fig. 7.10, Fig. 7.1 1,

and Fig. 7.12 respectively. The zero is placed at


-

12. It can be seen that the response

meets the requirements on overshoot and settling time.

page 117
Chapter 7 Controller Design Strategy

PD Controller for Linear TF model 1 1

Scope

53.50S+B47
*(T\ k> & th
s2+D.37BsrfQ0.Dfl54

Step Transfer Fen To Wottepace

Fig. 7.10 Transfer Function Model

Root Locus Open-Loop Bode Diagram

G.M: Inf
Freq: NaN
Stable loop

I;

;
t

PM: 78.4 deg


Freq: 56.7rad/sec

-30 -20
10 10 10
Real Axis
Frequency (rad/sec)

Fig. 7.1 1 SISO Output of the PD Controller

page 118
Chapter 7 Controller Design Strategy

Step Response
From i
1 4 r

System: Closed Loop: r to y


Peak amplitude: 1 .06

1.2
Overshoot (%): 21
At time: 0.0652

~~~~^
1 jr
' System: Closed Loop: r to y
^^^
I
/ Settling Time: 0.21 2

0.8 _
J System: Closed Loop, r to y .

DC 9BirT B7B
fl System Closed Loop: r to y
/
'

Is Rise Time: 0.022


05
/ 1 1 1

04 / ' ' ' -

0 2 ill 1 .

f 1 1 1

n 1 1 I 1 i i 1
0.05 0.1 0.1 0.2 0.25

Time (sec)

Fig. 7.12 Step Response of PD Controller

2. Lead Compensation Controller

Compensation with a transfer function of the form

s + z

C(s) = K (7.20)
s +p

is called lead compensation if z < p and lag compensation if z > p. Only lead

compensation is considered here because of the nature of the design requirements. Lead

compensation approximates the function of PD control and acts mainly to speed up a

decrease the transient The objective here


response by lowering rise time and to overshoot.

is to come with a pole-zero combination such that z < p and adjust K accordingly to
up
The trial and error.
meet the specifications. following values are arrived after

1 + 0.0671s s + 14.9\
C(s) = 808i 949501 (7.21)
1 + 0.000571s s + 1751 J

page 119
Chapter 7 Controller Design Strategy

where K =
94950, z = -

14.9, and p = -

1751. The closed loop transfer function

model, the outputs of the sisotool, and the response are shown in Fig. 7.13, Fig. 7.14, and

Fig. 7.15 respectively.

Linear TF model lead controller


k
w
I I
Scope!

s+14.0 1
h* th
w
s2+0.378s+Q0.0954

To WrjitepaceS
Transfer Fcn2

Fig. 7.13 Transfer Function Model

Root Locus Open-Loop B

400 '

\\
300
a

200
E
HI

100
I
GW:lnf

0 4 Freq:lnf
Stable loop
f
-1D0
/-

-200

1?
/3

i1
-300

-90

-400 \ i- P.M: 74 di
Freq: 57.6 rad/sec
i i i _ i v ^ 180
0 10' 10"
10

-2000 -15D0 -1000 -5D0


10 10
Real Axis
Frequency (red/sec)

Fig. 7.14 SISO Output of the Lead Controller

page 120
Chapter 7 Controller Design Strategy

Step Response
1.4 -

i
System: Closed Loop: r to y
Peek amplitude: 1 '<
.09

;
1.2 Overshoot (%): 21 .2
j i
At time: 0.0633

jf
~~

1 ;_ [ L^__^ System: Closed Loop : rtay


Settling Time: 0.179

0.8 System: Closed Loop: r to y


T!
3
/ System: Closed Loop: r to y ;
DC gain: 0.9

a. 1 (', 1 Rise Time: 0.021 2 I i i ii


^ 06 / >
' ' i i '
!
I
i / ! i
'
i
'

i
/ ;i : ii I I
04
i / :i : ;i ; I

02
\ li : ii ; i I

f i i li I I

a i\ ii i ii! i i I
0.02 0.04 0.06 0.08 0.1 0.12 0.14 0 16 0.1E

Time (sec)

Fig. 7.15 Step Response of Lead Compensation Controller

3. PID Controller

The third type of controller suitable for the case under study is the classical PID

(Proportional, Integral, and Derivative) controller. It is generally seen that proportional

(P) controller can reduce the error responses to disturbances, but it still can allow a

nonzero steady-state error to constant inputs. If the controller includes a term proportional

to the integral of the error (I), then the steady-state error can be eliminated, although it

may deteriorate the dynamic response. Addition of a term proportional to the derivative of

the error (D) can often improve the dynamic response. Combination of these three terms

form the classical PID controller. The complete triple controller is described by the

following transform equation:

kps2
kr + kps + ki
C(s) =
kp H h kns (7.22)
s

The following values are arrived after trial and error:

page 121
Chapter 7 Controller Design Strategy

2500(1 + 0.143s) (1 + 0.0833s)


C(s) -,

which after simplification becomes:

29.78s2
+ 565.61s + 2499.02
C(s) =
(7.23)

Therefore the proportionality constants are: kr> =


29.78, kp =
565.61, and

fcj =
2499.02. The closed loop transfer function model, the outputs of the sisotool, and

the response characteristics are shown in Fig. 7.16, Fig. 7.17, and Fig. 7.13 respectively.

Linear TF model PD controller


I I
h
W

Scope4

1
+Q- PID h
w th
gZ+0 .3708+90.0054

Step4 PID Controller To Workspaces


Transfer Fon5

Fig. 7.16 Transfer Function Model

page 122
Chapter 7 Controller Design Strategy

Root Locus Open-Loop Bode Diagram

G.M.: Inf
Freq: NaN
Stable loop

! !
^j

P.M.: 60.2 deg W^


Freq: 34.7 rad/sec

-40 -30 -20


10 10
Real Axis
Frequency (rad/sec)

Fig. 7.17 SISO Output of the PID Controller

Step Response
System: C losed Loo >:rtov
From:U(1)
1.4 : I I i
Peak amp itude: 1 6 .1

Overshoot(%):1B.4
At time: 0 0896
1.2

i SVstem: ClosBcl LP: <*


/ii\ i
'
^ item: Closed Loop: r to y
DC 9aln: 1
1
Settling Time: 0.549 - _
\ ,

V LL~. iZ. V _LV lil V _ _

T i: x. :

[I System: Closed Loop: rto y ; i :


0.3 r
0) &
f Rise Time: 0.03 79
j\
2 1
i>
i I I i
0.8

! ! i
i
0.4
-j--.--"-,';
'
I 'i
0.2 Ll.jj

I l! 1 i i ! i
0.1 0.2 0.3 04 0.5 0 09

Time (sec)

Fig. 7.18 Step Response of PID Controller

page 123
Chapter 7 Controller Design Strategy

A comparison of the three controllers is shown in the Fig. 7.19. Although,


following all

the three controllers meet the overshoot and


settling time criteria, only the PID controller
gives unity DC gain.

Step Response
1 4 1 1 1 -
i r

PD Controller
Lead Compensator
I 2 PID Controller

--\ i t r
11 i v-^4_ J-^T
0.8 -

11 '

0.6
MM
0 4
IMM!
0.2
M MM 1
i 1 ! i !
0 0.1 0 2 04 0.5 0.6 07 0.8 0.9

Time (sec)

Fig. 7.19 Comparison of PD, Lead, and PID Controllers

As the next step towards modifying these controllers for the nonlinear system, the step

input is replaced with the sinusoidal runway profile and all the three controllers are used

with derivative feedback in order to stabilize the response. A pre-filter is used to

smoothen the input signal. The modified controllers for the nonlinear system and their

performances are given below:

1 . Proportional Feedforward and Derivative Feedback

After trial and error, we arrive at: kp =


2, and kn = 5. The block diagram representation

of the controller with the nonlinear model and the response of the system are shown in

Fig. 7.20 and Fig. 7.21.

page 124
Chapter 7 Controller Design Strategy

Proportioanal Feedforward and Derivative Feedback

?
Scope3

F\7
Sine Wave2
pre.(i|te, x
R> >
To Wodspacel

&
850"
D -B53"sin(ij [1 J>0 co<u [1 J)

Fig. 7.20 Proportional Feedforward and Derivative Feedback

Proportional Feedforward and Derivative Feedback


0.045 1 1 i

0.Q4

0.035

0.03
%Mp== 13.19
ts =
l.jffi sec
S 0.025
O
S.S =
0.0379
CL

| 0.02

0.015 H

0.01

0.005

0 i i i
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Time

Fig 7.21 Controller Performance on the Nonlinear System

1. Lead Feedforward and Derivative Feedback

The tested values are kP =


3, kD 4, z =
-1.5, and p = -

2. The block diagram

representation of the controller and the response of the nonlinear system are shown in Fig.

7.22 and Fig. 7.23.

page 125
Chapter 7 Controller Design Strategy

Lead Controller and Derivative Feedback

? ?

h
Sine Wavel Pre.fl|tert

^Im^H 9+1.5

9+2

To Wortepaee2

PI Controlled

3 n(u [1 ]>0
.853-si cos(u [1 p

Fig. 7.22 Lead Feedforward and Derivative Feedback

Lead Feedforward and Derivative Feedback


0.045 I i

0.04

0.035

D.03
%Mp =
13.19
ts= 1.{!5 sec
S 0.025
S.S = 0.0379
?

tfi
0.02

0.015

0.01

D.0Q5

i i
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Time

Fig. 7.23 Controller Performance on the Nonlinear System

1. Proportional, Integral Feedforward and Derivative Feedback

The proportionality constants obtained after trial and error are: kP 1.2, kT =
0.1, and

8. The block diagram of the controller and the response of the


kD = representation

nonlinear system are shown in Fig. 7.24 and Fig. 7.25.

page 126
Chapter 7 Controller Design Strategy

Scope! Scope2

Scope4
N si_ Scopel
Runway profile pre-filtor

? "J- PD In th th1

PI Controller! Plant Model2 To WorHspace2


Product2

PD

PI Controlled

0.853^ln(U[1])-0.0656"COS(u(1D *

Fcn2

timet
Ctoc* To WorHspacel

Fig. 7.24 Proportional, Integral Feedforward, and Derivative Feedback

Proportional, Integral Feedforward and Derivative Feedback


0.045 I 1 1

0.04

0.D35

0.03
%Mp 1=13.72
ts= 1.{>5 sec
0.025
SS =
0.0376 j
0.02

0.015

0.01

0.005

' i i
0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Time

Fig. 7.25 Controller Performance on the Nonlinear System

A comparison of the three controller performances is shown in the following Fig. 7.26.

page 127
Chapter 7 Controller Design Strategy

Comparison of the Three Controllers


0.045 i
; i i f
; ;

0.04
^'
\
I
0.035
Jn *~^~\

0.03
Proportional Feedforward
Lead Feedfoiward
0.025
Proportional, Integral Feedforward

0.02

0.015

0.01

0.005

i i i 1 i i i
0.5 1 1.5 2 2.5 3 3.5 4.5
Time

Fig. 7.26 Comparison of the Three Controller Performances

The controller design approach described above is only one of the many available

methods of design. In fact, for more complex multi-degree of freedom systems, the trial

and error method will be a lengthy process before we actually get a convergence of the

solutions. In the next chapter, a more sophisticated approach is taken for designing a

controller for the 3-dof model. This approach makes use of the state space design

techniques and full state feedbacks with the reference input. The design ensures

minimum steady state error to the reference input by choosing the feedback gain values

and the input scaling factor.

page 128
Chapter 8

CONTROL OF THE MEMO MODEL

8.1 Introduction

The control of a Single Input Single Output (SISO) system, described in the previous

chapter is straight forward and easy compared to the control of a Multi Input Multi
Output (MTMO) system. The reason is quite obvious that a MTMO system has multiple

control inputs, which produce multiple responses and these responses, as seen in most

cases, could be coupled in the different coordinates of the system [19]. In this chapter, we

will follow the state-space approach of designing a controller because it is more suitable

to complex systems with multiple inputs and multiple outputs. In state-space design, a

dynamic compensation for the system is designed by working directly with the state-

variable description of the system. A state variable description of a system is an

alternative way to represent the dynamic equations of motion. The differential equations

describing a dynamic system are organized as a set of first-order differential equations in

the vector-
valued state of the system, and the solution is visualized as a trajectory of this

state vector in space. For example, a second order system can be equivalently represented

by two first order differential equations. This requires two state variables, the rate of

change of each state variable represents a single differential equation.

As an example, let us consider the equation of motion derived in Chapter 4 (Eqn. 4.23)
for the 1-dof model without considering the input runway profile:

9 + 90.0234 9 =
3.415 (8.1)

Since the order of the ODE is 2, we require two state variables to represent the dynamics

of the system in state variable format. Let us define two state variables xx, and x2 such

that we have:

9 xx; and 9 = =
=
xx x2,

x2= 90.0234 xx + 3.415.


-

These three equations can be rearranged in matrix form as:

page 129
Chapter 8 Control of the MIMO Model

x =
Ax + Bu,
(8.2)

=
Cx + Du
y (8.3)

where:

0 1
A =
B
90.0234 0
=
(8.4a)

1 0
C =
D =
0, u= [3.415]. (8.4b)
0 1

In the state variable formulation A, B, C, and D represent the matrices of coefficients

and

x =
vector of state variables

u =
vector of inputs

y =
vector of outputs

Thus, an nth-order differential equation would be represented by n first-order differential

equations, and the system responses can be obtained by the single integration of n

simultaneous equations rather than n integrations of one equation. The initial conditions

on 9, and 9 can be represented in the sate space form as:

ari(0)
x(0) (8.5)
x2(0)

In general, if the system is represented by an nth order differential equation with m

inputs; x will be an nth order vector, u will be an mfh order vector, A will be an (n *
n)

matrix, and B will be an (n x


rn) matrix. If k output variables are to be monitored, y will

be a fcth-order vector, C will be (k x


n), and D will be (k x
m) [9].

page 130
Chapter 8 Control of the MIMO Model

8.2 State Space Approach to Controller Design

Even though state space design techniques can be applied to nonlinear systems as well,

the analysis becomes complicated unless we know the nonlinear dynamics present in the

system. So it is always better to study the linearized version of the nonlinear system and

study the influence of the control parameters on the system. So in the next step we will

find out the equilibrium conditions of the system and do a linearization of the nonlinear

3-dof model about these equilibrium conditions.

8.2.1 Static Equilibrium Conditions

The nonlinear differential equations representing the dynamic equations of motion of the

3-dof aircraft landing gear model are reproduced below in input-output separate form:

Equation of Motion 1:

6=(z1 + zQ) (0.853 sin9


-

0.065 cos9) + zx (47.353 9 -

437.37) (8.6)
+ z2(A37.37
-

47.353 9) + (3.892 -

125.43 9)
9 (0.005 9
cos2

+ cx sin9 cos9
-

0.023 -

0.000275)
+ cx(zx -

z2) 0.00834 sin9


-

0.077 cos9

Equation of Motion 2:

zx = 0(0.1647 -

sin9
-

0.013 cos9) + 277.07(z2 -

zx) (8.7)
[0.1647 92
(9)2

+ cos9 + 0.013 sin9\ + (4.57 -

84.45 9 -

6.94)
+ cx (0.0016 sin9
-

0.015 cos9) + 0.0488 cx(z2


-

zx)
-

0.998 zq + 0.0843 km(z0 -

zx)

Equation of Motion 3:

z2= -

1109.31 + 106012.07(21
-

z2) (8.8)
+ (32312.48 9 -
1749.2 92) + 32.258 kn(zH -

z2)

0.616 sin9) + 18.675


-

+ cx f9(5.692 cos9
-

cx(zx z2)

page 131
Chapter 8 Control of the MIMO Model

To linearize these equations of motion, we need to first determine the appropriate

equilibrium conditions about which the linearization can be performed. Before we

proceed, we will define a state vector for the 3-dof model as follows:

9 =
xx; 9 =
xx =
xtz, 9 =
4;
zx =
x2; zx =
2 =
5; zx =
x5;
z2 =
x3; z2 =
3 =
x6; z2 =
6;

The set of equations can be expressed in vector for as:

~xx(t)~ '9'

x2(t) z\

xz(t) z2
x(t)
9
(8.9)
xA(i)
x5(t) zx

_x6(t)_ ,'zi.

To determine the equilibrium conditions on the state variables, we refer to Chapter 3,


where we did the static analysis of the system. The static analysis of the system showed

that the equilibrium condition on the angle cf) =


24.67 deg =
0.4306 rad. Now, we can

derive the equilibrium condition on 9 from the expression of coscj) as follows:

(QF cos9 + QP sin9


-

PD cos9)
cos(4>
9) ED
(8.10)

Substituting for the known parameters, we obtain:

1.765
9e= = 0.0308 rad.

But a major assumption we made in this derivation is that zx and z2 are equal to zero.

Also, for the approximated version of coscj), i.e.

(QF cos9 + QP sin9


-

PD cos9)
cos(cj)) (8.11)
ED

1.657
9P= = 0.0289 rad.

page 132
Chapter 8 Control of the MIMO Model

It can be seen that the static equilibrium value of 9 is different from zero and is slightly
positive. This is because the center of
gravity of the plane is located at the rear of the
plane and this weight is causing the nose of the plane to be tilted upwards when the plane
is motionless. The static equilibrium condition of the plane is shown in Fig. 8.1. In order
to bring the angle 9 to zero when the plane is idle, the center of of the plane or the
gravity
distance QG has to be 0.08 m instead of 0.0127 m. This is calculated from Eqn. 3.3 and

Eqn. 3.4 by substituting the value of 0 corresponding to 9 =


0.

/
>i //////////// /^/

Fig. 8.1 Static Equilibrium of the Plane

Since we only have one equation relating the three coordinates 9, zx, and z2, we can not

find the values of zx and z2 corresponding to the equilibrium condition on 9. Moreover,

for a given input, the inherent dynamics of the system make it settle down to different sets

of steady state values. These steady state values can be obtained by running the
simulation of the nonlinear 3-dof model over a large period of time. By running the
simulation for 90 sec, we arrive at the following equilibrium conditions on the state

variables.

page 133
Chapter 8 Control of the MIMO Model

xx(t) 0.08025
x2(t) 0.0144
zq
x3(t) 0.000185
xe(t) =

x4(t) 0.00003905 ue(t) =


zq (8.12)
ZH
x5(t) 0.000116
xe(t) 0.000789

where the inputs are:

zq =
0.0025 sm(62.83 t) (8.13)
z0 = -

9.87 sm(62.83 t)

zH = 0.0025 sinl -

10 t -

0.033 sin9 + 0.3048 cos9

0.165 cos9 + 0.3048 sin9


-

0.132 cos9 + zx -

z2
-

0.046 Wl
0.046

where zh is the Eqn. 5.25 derived in Chapter 5, modified to include the effects of the

displacements zx and z2 from the equation of coscj) given by Eqn. 5.13.

The simulation is performed with the following initial conditions on the state variables:

"0.0873"

[an (0)1
x2(0) 0
a*(0) 0
Xi(0) 0
(8.14)
x4(0)
ar5(0) 0
0
X6(0)\

8.2.2 Linearization of the Nonlinear 3-dof Model

We can perform the linearization of the nonlinear 3-dof model about the equilibrium

conditions just obtained. To do this, we define input-output points in our Simulink block

diagram as shown in Fig. 8.1. The important thing to be considered here is, even though

we are giving only one input to the ODE of zx, the derivatives of the inputs are occurring

in ODE's for So have to define input points for 9, and z2 as well. Therefore,
9, and z2. we

The equilibrium values of the states can be


we have three input and three output points.

LTI tool box.


specified in the user defined Set Operating Point window of the viewer

page 134
Chapter 8 Control of the MIMO Model

Linearization of the 3-dof Nonlinear Model

-KID
th -?*

fV
zOdotdot model
vi-

2Dd
zOdotdot
21 zl'

-KD
z2 Z21

h
zD model
thdot
thdof

-KD
zldot
zldot
zh V4-
lh;1 z2 -KD
z2dot >z2dot z2dor

zh model Plant Model

Fig. 8.2 Linearization of the Nonlinear 3-dof Model

Once the linearization is performed, the state variable description of the linearized system

can be inherited by exporting the LTI viewer model into the Matlab workspace. The state

variable description of the system gives the coefficient matrices A, B, C, and D. The

three inputs to the system are zq in the ODE for zx, zq in the ODE for 9, and zh in the

ODE for z2. There are six states and three outputs for monitoring purposes. The output of

the Simulink linearized system is shown below:

a =

th Z1 z2 thdot zldot Z2d0t

th 4.8193e-021 0 0 1 0 0
Z1 0 0 0 0 1 0
z2 0 0 0 0 0 1

thdot -140.16 -436.42 434.4 -0.45926 -1 .5259 1 . 5247


zldot -85.814 -951 .58 277 . 18 -0.29662 -1.3979 0 97637
.

z2dot 32077 1 .0614e+005 -3 6422e+005 112.49 373.5 -373.5

b =

zO zOd zh

th 0 0 0
z1 0 0 0
z2 0 0 0
thdot 2.0182 5 9853e- 006 1 2735e-013

zldot 674.4 -0. 998 7.797e-014

z2dot 0 0 2 5806e+005

page 135
Chapter 8 Control of the MIMO Model

th Z1 Z2 thdot zldot z2dot


th 1 0 -2.2337e-015 3 8462e-020 -1 .5385e-019 0
Z1 0 1 5.4729e-016 6, ,9792e-021 -2 .7917e-020 0
z2 2.0675e-018 0 1 -3.,5628e-020 1 ,4259e-019 0

d =

zO zOd zh
th 0 0 -9.0859e-016

z1 0 0 1.6262e-016
z2 0 0 0

Continuous-time model.

Now that we have obtained the matrix representation of the dynamics of the linearized

system, we can run a linear simulation using the sinusoidal inputs introduced earlier to

find the responses of the linearized system. A listing of the Matlab codes for rmming a

linear simulation with the system matrices is shown following the state variable

description of the system.

A=[4.8193e-021, 0, 0, 1, 0, 0;

0, 0, 0, 0, 1, 0;
0, 0, 0, 0, 0, 1;
-140.16, 434.4, -0.45926, -1.5259, 1.5247;
-436.42,

-85.814, 277.18, -0.29662, -1.3979, 0.97637;


-951.58,

32077, 1.0614e+005, -3.6422e+005, 112.49, 373.5, -373.5];

B=[0, 0, 0;
0, 0, 0;
0, 0, 0;
2.0182, 5.9853e-006, 1.2735e-013;
674.4, -0.998, 7.797e-014;
0, 0, 2.5806e+005];

C=[1, 0, -2.2337e-015, 3.8462e-020, 0; -1 .5385e-019,

0, 1, 5.4729e-016, 6.9792e-021, -2.7917e-020, 0;


2.0675e-018, 0, 1, -3.5628e-020, 1.4259e-019, 0];

D=[0, 0, -9.0859e-016;
0, 0, 1.6262e-016;
0, 0, 0];
'thdot' 'zldot'
' 'z2'
states =
{'th' '
z1 'z2dot'};
{'zO' 'zOd'
inputs =
'zh'};
{'th' '
outputs = 'z1 'z2'};
'
sysjnimo =
ss(A,B,C,D, statename ', states, .. .

'inputname'

,inputs, . . .

'outputname'

.outputs)

t=0:0. 01:90;
x0=[5*pi/180 0 0 0 0 0];
u1=a*sin(omega*t);
u2=-a*omega~2*sin(omega*t);

PD*cos(y1)).'2)));
u=[u1', u2', u3];

[y,x]=lsiin(A,B,C,D,u,t,xO);

page 136
Chapter 8 Control of the MIMO Model

yi=y(:,D;
y2=y(:,2);
y3=y(:,3);

figure (1);
subplot (3,1 l1);plot(time,thltIy1);grid;
title ( 'Nonlinear vs. Linear Responses');
'Nonlinear'
legend ( , 'Linear');
ylabel( 'Response of Th (rad)');
subplot (3 , 1,2); plot ( time, z1,t,y2); grid;

ylabel( 'Response of z1 (m)');


subplot(3,1,3);plot(time,z2,t,y3);grid;
ylabel( 'Response of z2 (m)');

Nonlinear vs. Linear Responses

0.02

mmmm
mmmm
-0.02

40 50 60 70 80 90

0.01

CN

-0.01

cc
-0.02

10 20 30 40 50 60 70 80 90

Fig. 8.3 Nonlinear vs. Linear Responses

A comparison of the nonlinear and linear responses (Fig. 8.3) shows that the linear model

is a good approximation in representing the dynamics present in the system except for the

state values. This indicates that the method of linearization has taken care of most
steady
aspects of the dynamics of the original nonlinear model.

page 137
Chapter 8 Control of the MIMO Model

The transfer function models are given below and the root locus plots of the transfer

function models are shown in Fig. 8.3.

"zO"
Transfer function from input to output...

2.018 S"4 -
272.4 S~3 + 4.427e005 S'2 2.648e008 s 7.546e010
-

:h: - --

s"6 + 375.4 s'5 + 3.655e005 s"4 + 7.796e005 s'3 + 3.544e008 sA2 + 9.879e007 s + 2.539e010

674.4 s"4 + 2.522e005 s"3 + 2.457e008 S~2 + 8.202e007 s + 2.499e010


i\ : - -

SA6 + 375.4 s"5 + 3.655e005 s'4 + 7.796e005 S~3 + 3.544e008 S~2 + 9.879e007 S + 2.539e010

2.521e005 S~3 + 7.165e007 s~2 + 2.24e006 s + 6.35e008


z2: -

S'6 + 375.4 s~5 + 3.655e005 s~4 + 7.796e005 S"3 + 3.544e008 SA2 + 9.879e007 s + 2.539e010

"zOd"
Transfer function from input to output...

5.985e-006 s"4 + 1.525 S"3 + 438.2 s"2 + 3.939e005 s + 1.126e008


th: - -

S'6 + 375.4 s"5 + 3.655e005 s"4 + 7.796e005 s"3 + 3.544e008 sA2 + 9.879e007 s + 2.539e010

-0.998 s"4 373.2 S'3 3.636e005 s~2 1.216e005 s 3.704e007


- - -

i\ : - - -
-

SA6 + 375.4 sA5 + 3.655e005 s~4 + 7.796e005 SA3 + 3.544e008 S~2 + 9.879e007 s + 2.539e010

-372.8 S"3 1.059e005 SA2 3050 S 8.757e005


--
-

'.2; --

s"6 + 375.4 S~5 + 3.655e005 S~4 + 7.796e005 s'3 + 3.544e008 sA2 + 9.879e007 s + 2.539e010

"zh"

Transfer function from input to output...

-5.729e-010 sA4 + 3.935e005 sA3 + 1.123e008 s~2 + 3.12e008 s + 7.546e010


-

h: -

+ 3.655e005 s~4 + 7.796e005 s'3 + 3.544e008 sA2 + 9.879e007 s + 2.539e010


sA6 + 375.4 s"5

page 138
Chapter 8 Control of the MIMO Model

2.52e005 s'3 + 7.153e007 s'2 + 1.149e006 s + 4.057e008


zl:

SA6 + 375.4 S'5 + 3.655e005 S"4 + 7.796e005 s"3 + 3.544e008 s'2 + 9.879e007 s + 2.539e010

2.581e005 s"4 + 4.793e005 S"3 + 2.818e008 sA2 + 9.614e007 s + 2.475e010


z2:

SA6 + 375.4 SA5 + 3.655e005 SA4 + 7.796e005 SA3 + 3.544e008 sA2 + 9.879e007 s + 2.539e010

Root Locus
q6

x 1 Root Locus

2000
2000

1000 1000
n
0 -& 3 O i o -0
K
Q--\
Z\.
-1000
a -1000

-2000

:k
-200 0 200 400 -1 o 1 -2000 -1000 0 1000

Real Axis Real Axis Real Axis


x -)

Root Locus Root Locus Root Locus

-
500 2000
:/
; A
500 .4_.
1000

0 J i J :--_
<y
3. A 0 <S4 :
cri

E -1000
...*V-._

-500 ;$$-->
; \
-2000
i N
-1000
.500 : i\
<a :

-50 0 -200 0 200 -2000 -1000 0 1000


-200 -150 -100

Real Axis Real Axis


Real Axis

Root Locus Root Locus


Root Locus

; 1 000
2000
/. 2000
i / 500
1000 1000
to m

ifSA ...
0 0 y-\-~\--\-4-
0 %3r
en
x. \ ', !
i
-1000
S -1000 -500 TV 1
1*
1 T

-2000
-2000 -1000
/
', 0
-2000 -1000 0 1000 -1000 0 1000 2000 -200 -150 -100 -50

Real Axis Real Axis


Real Axis

Fig. 8.4 Root Locus Plots of the Transfer Function Models

poles of the characteristic


equation of the system are given below:
The eigenvalues or the

page 139
Chapter 8 Control of the MIMO Model

eig(A)

=
ans

-1.871 le+002 +5.7364e+002i


-1.871 le+002 -5.7364e+002i

-6.7782e-002 +8.8269e+000i
-6.7782e-002 -8.8269e+000i

-5.0253e-001 +2.9913e+001i
-5.0253e-001 -2.9913e+001i

A close observation of the root locus of the transfer function of 9 shows that there is an

unstable transfer function zero in the right half of the complex plane, which can be found
as:

tzero(GLTCP(l,l))

=
ans

-8.5653e+001 +6.0409e+002i
-8.5653e+001 -6.0409e+002i

5.0514e+002 (unstable zero)


-1.9884e+002

The presence of the right half plane zero makes the system difficult to control. However,
we can examine the controllability and observability of the system associated with the

control input u(t) and the system outputs X{(t).

In general, a nth-order system is controllable if and only if the controllability matrix

P= [B AB A2B ...
An~1B] =
ctrb(A,B), has full rank n (where nis the

order of the state space model). If the rank of P is k < n, then the system has n k

uncontrollable modes [20].

Similarly, a nth-order system is observable if and only if the observability matrix

C
CA
Q =
CA2
=
obsv(A, C), has full rank n. If the rank of Q is k < n, then the

CAn-\
system has n k unobservable modes.

To determine which modes are uncontrollable or unobservable, we can make use of the

modal transformation T. Let T be the matrix of eigenvectors of the system matrix A such

that we have:

page 140
Chapter 8 Control ofthe MIMO Model

[T,A] =
eig(A)

If we represent the eigen vectors of the system matrix by C, the state vector becomes

x(t) =
TC(t). This transformation will transform the state space description of the

system as described below:

x(t) =
Ax(t) + Bu(t) (8.15)
y(t) =
Cx(t) + Du(t)

Substituting for x(t) we get:

C(t) =
T-lATl(t) + T^BuW (8.16)
y(t)=
CT{(t) + Du(t)

The new coefficient matrices are represented by:

A-modal T AT , Bmodal T B (8.17)


Cmodal CT j "modal -tJ

The transformation matrix T renders the system matrix A into the modal form which is

diagonal in nature. The generic form of Amodai is given by:

Ai 0 0
0 A2 0
Amodal T~lAT = A (8.18)
0 0 An

When the system is in modal form, the system controllability can be determined by
checking whether any rows of the matrix B^dai are zeros. In other words, the mode A; is
uncontrollable if and only if the row Bmi = 0.

Similarly, the observability of the system can be determined by checking whether any

columns of the matrix Cmodal are zeros. In other words, the mode A* is unobservable if

page 141
Chapter 8 Control of the MIMO Model

and only if the column Cv 0. The entire procedure can be performed using the
following Matlab codes:

T=eig(A);
'modal'
[Am,Bm,Cm,Dm,T] =
canon(A,B,C,D, )

Am =

-1 .8711e+002 5 ,7364e+002 0 0 0 0
-5.7364e+002 -1 ,8711e+002 0 0 0 0
0 0 -6.7782e-002 8.8269e+000 0 0
0 0 -8.8269e+000 -6.7782e-002 0 0
0 0 0 0 -5
,0253e-
001 2.9913e+001
0 0 0 0 -2 ,9913e+001 -5.0253e-001

Bm =

1 . 9724e+002 -2.9162e-001 -2.5798e+005

-3.7639e+002 5.5649e-001 8.3879e+004


-2.5680e+002 3.8296e-001 2. .2692e+002

5 1 832e+000
. -7.6906e-003 -5..5431e+000

7.3588e+002 -1.0887e+000 2. .4077e+002

-1.0421e+001 1 .5407e-002 2.,2535e+001

Cm =

1 . 7875e 006
- -
3 6294e 006
.
-

-8.5997e-004 -1.1199e-001 3.2138e-004 -1.1651e-002

1.1493e-006 -2.3259e-006 -4.5488e-005 8.6743e-003 -4.9969e-004 -2.9744e-002

5.1393e-004 1.5756e-003 7.1428e-005 -7.3388e-003 6.0688e-004 -9.7368e-003

Dm =

0 0 -9.0859e-016

0 0 1 .6262e-016

0 0 0

-4.0665e-002 -6.1488e-001 -2.9927e-001 8. 8091 e-


002 2.9220e-001 -9.9968e-001

-5.5979e+001 -1.8572e+002 6.3484e+002 -1.6760e-001 -5.5761e-001 3.2504e-001


1.1181e-001 7.2759e-001 7.9438e-001 9.8198e-001 -3.8372e-001 8.7934e-004
-8.6657e+000 3.3950e+000 1.2965e-002 -6.7831e-003 7.7060e-003 -2.1480e-005

6.8876e-002 2.9553e-001 8.4866e-001 8.4563e-002 1 .0909e+000 9. 3301 e-


004
-3.2621e+001 -3.7962e-002 -4.8839e-003 -1 .5438e-002 -8.7326e-005
-2.5241e+000

Since no rows of Bm or no columns of Cm are equal to zero, we can conclude that the

system is controllable and observable.

page 142
Chapter 8 Control of the MIMO Model

8.3 Full State Feedback Control Design

In this section, we will study the design of a control law for full state feedback control.

The full state feedback assumes that all the states are available for measurement and

feedback. However this may not be true in practice, because in many cases we can not

measure all the states and make them available for feedback. This case has to be dealt

with separately by considering the estimates of the actual states of the system. The block

diagram representation of full state feedback is shown in Fig. 8.5.

x=Ax+Bu

y=Cx+Du

u=-Kx

Fig. 8.5 Block Diagram Representation of Full State Feedback

8.3.1 Control Law Design for Full State Feedback

the law is to allow us the flexibility to assign a set of pole


The purpose of control

that will correspond to a satisfactory response in


locations for the closed loop system

other measures of the transient response.


terms of the rise time and

form follows:
the linearized 3-dof model in state space as
Let us consider

page 143
Chapter 8 Control of the MIMO Model

4.8193e-021 0 0 1 0 0
0 0 0 0 1 0
0 0 0 0 0 1
A =

-140.16 -436.42 434.4 -0.45926 1.5259 1.5247 (8.19)


-85.814 -951.58 277.18 -0.29662 1.3979 0.97637
32077 1.0614e+005 -3.6422e+005 1 12.49 373.5 -373.5

0
0
0
B =

2.0182 5.9853e-006 1.2735e-013


674.4 -0.998 7.797e-014
0 0 2.5806e+005
1 0 -2.2337e-015 3.8462e-020 -1.5385e-019 0
C =
0 1 5.4729e-016 6.9792e-021 -2.7917e-020 0
2.0675e-018 0 1 -3.5628e-020 1.4259e-019 0
0 0 -9.0859e-016

D 0 0 1.6262e-016
0 0 0

In this model, the state vector


x(t) consists of the following state variables:

Pitch angle 9(t) =


xx(t)

Displacement of the center of the main wheel


zx(t) =
x2(t)

Displacement of the center of the nose wheel


z2(t)
=
x3(t)

Pitch angle rate 9(t) =


x^(t)

Velocity of the center of the main wheel zx(t)


=
x$(t)

Velocity of the center of the nose wheel z2(t)


=
x^(t)

and the state equations are:

x(t)
=
Ax(t) + Bu(t) (8.20)
y(t)
=
Cx(t) + Du(t)

The full state control feedback law is to find feedback gains K such that the control law

is of the form:

page 144
Chapter 8 Control of the MIMO Model

xx
x2
X3
u(t)= -Kx(t)=
-\KX K2 Kz KA K5 K6] (8.21)
X4
x5
xq

As we have six states in our system, we have six feedback gains at our disposal to be
determined. There are two approaches in the design of the full state feedback gain matrix

K:

(1) Eigen assignment or pole placement method

(2) optimal control based on linear quadratic regulator (LQR) method

8.3.2 Eigen Assignment or Pole Placement Method

Given that the system is controllable, the full state feedback gain matrix K is determined

such that the eigenvalues of the closed loop system are at the desired pole locations.

Substituting the feedback law in Eqn. 8.20, we obtain:

x(t)
=
Ax(t) -

BKx(t) (8.22)

Therefore, the characteristic equation of this closed loop system is:

det si
-(A-

BK) (8.23)

When evaluated, this will yield a 6 th-order polynomial in s containing the gains K{. The

control law design then requires picking the gains Ki so that the roots of the characteristic

equation are in desired locations. Suppose the desired closed loop pole locations are

the loop the desired


^desired =
{pi, P2, ,
P&] ^ order to place closed system poles at

pole locations we should have:

detlsl -(A-

BK)\
=(s-pi)(s-p2)...(s-

pe) (8.24)

page 145
Chapter 8 Control of the MIMO Model

This forces the system's characteristic equation to be identical to the desired characteristic

equation and the closed loop poles to be placed at the desired locations. The major

difficulty in this approach is that it takes a great amount of practice and trial and error to

come up with a desired characteristic equation which will ensure the performance

requirements. One way is to utilize Matlab's sisotool feature to design the desired pole

locations. Since the plant state space model is of 6th-order, we pick six pole locations to

design the feedback gains. The following six pole locations are picked at random and

manipulated by trial and error based on the performance requirements discussed in

Chapter 7, which were basically a 1 % settling time of not more than 2 sec and a

percentage overshoot not greater than 25 %.

Pi =
-10 (8.25)
=
Pi -15

P3 = -

40 + AOi
P4 = -

40 -

40i

Pb = -

100 + lOOi

P6 = -

100 -

lOOi

The root locus plot and the step response corresponding to the above set of pole locations

are shown in Fig. 8.5 and Fig. 8.7.

Open-Loop Beds Dsscram

10'

-200 0 10
Real Axis Frequency (rad/sec]

Fig. 8.6 Root Locus Plot of the Desired Pole Locations

page 146
Chapter 8 Control of the MIMO Model

Step Response
0.7 1
System: Closed Loop: 1
r to y
Peak amplitude: 0 565
Overshoot (%): 13
0.6 -

"At time: 0.264 '

-
^^~. ,, System. Closes Loop: r ro y
y I '^""--^^ Settling Time: 0.389
0.5 hv.
I : i System: Closed Loop: r to y
DC 0.5
1 i (\ Sy stem: Closed Loop: r to y
gain:

0.4
ieTime: 0.107
i

1
03

0 2
1

01 lZl J... J.

y \ ! i I 1 !
0.1 0.2 0.3 0.4 0 S 0 6 0.7

Time (sec)

Fig. 8.7 Step Response of the Chosen Pole Locations

Once we know the desired pole locations of the closed loop system, we can use Matlab's

place command to find the gains Kz.

8.3.3 Introducing the Reference Input

In order to study the transient response of the pole placement designs to input commands

it is necessary to introduce the reference input r(t) into the system. This modifies the

control law as:

u Kx + r (8.26)

have nonzero steady state error to a reference input.


However, the system wil a nearly
state values of the state and the control
This can be corrected by computing the steady

input that will result in zero output error. If the desired final values of the state and the

law becomes:
control input are xss and uss, then the new control

u =
uss
-
K(x -

xss) (8.27)

page 147
Chapter 8 Control of the MIMO Model

When x =
xss (no error), u uss. To apply this new control law to our state space model

given by Eqn. 8.20, we consider the case in steady state as:

0 =
Axss Buss
+ (8.28)
Uss =
Cxss + Duss

In order for the steady state value of the output to be equal to the steady state value of the

reference input for zero steady state error, we should have yss = rss. To do this we make

the substitutions xss =


Nxrss and uss Nurss, which implicitly means we are scaling
the input to match the steady state values of the state and the control input. With these

substitutions Eqn. 8.28 can be written in the following matrix form:

'Nx' '0'

'A
C D
=

1
(8.29)

This equation can be solved for Nx and Nu to get:

-i

'A "0
=

_NU_
C D 1

With these values of Nx and Nu we can modify the control law given by Eqn. 8.27 for

to a step input:
introducing the reference input to get zero steady state error

u = Nur -

K(x -

Nxr) (8.30)
=
-Kx+(Nu + KNx)r

The coefficient of r in the parentheses is a constant, which is normally represented by N.


Therefore we have:

u= -Kx + Nr. (8.31)

defined function called rscale [21] can be created in Matlab to find the input gain
A user

N. The Matlab codes to calculate N are given below:

function [Nbar]=rscale (A, B,C,D,K)

page 148
Chapter 8 Control of the MIMO Model

% |Nx| |A B|-1 10J


% 1 1 =
1 I I I
% |Nu| IC D| Ml
%

=
s size(A,1); % To find the size of A matrix
[row, col] =
size(B); % To find tf)e size Qf B matrix
Z -

[zeros([1,s]) ones(1 ,col) ]; % To create a column vector


N
=_inv([A,B;C,D])*Z'; % To find the inverse
Nx -

N(1:s); % Scaling factors for the states


Nu -

N(1+s); % scaling factors for the control inputs

Nbar=Nu +
K*Nx; % Combined
scaling factor for the input

A linear simulation utilizing pole placement design to calculate the feedback gains is
carried out with
step inputs in place of the sinusoidal inputs.

% Codes to Evaluate the Feedback Gains for the States

% State Space Model in Generic Form

A=[4.8193e-021, 0, 0, 1, 0, 0;
0, 0, 0, 0, 1, 0;
0, 0, 0, 0, 0, 1;
-140.16, 434.4, -0.45926, -1.5259, 1.5247;
-436.42,
-85.814, 277.18, -0.29662, -1.3979, 0.97637;
-951.58,

32077, 1.0614e+005, -3.6422e+005, 112.49, 373.5, -373.5];

B=[0, 0, 0;
0, 0, 0;
0, 0, 0;
2.0182, 5.9853e-006, 1.2735e-013;
674.4, -0.998, 7.797e-014;
0, 0, 2.5806e+005];

C=[1, 0, -2.2337e-015, 3.8462e-020, .5385e-019, 0; -1

0, 1, 5.4729e-016, 6.9792e-021, -2.7917e-020, 0;


2.0675e-018, 0, 1, -3.5628e-020, 1.4259e-019, 0] ;

D=[0, 0, -9.0859e-016;
0, 0, 1.6262e-016;
0, 0, 0];

B1=B(:,1) % First Column of B matrix


Corresponding to Input zO
B2=B(: ,2)
% Second Column of B Corresponding to Input zO
matrix

B3=B(:,3) % Third Column of B matrix Corresponding to Input zO

C1=C(1,:) % First Row of C matrix Corresponding to Input zO


D1=D(:,1) % First Column of D matrix Corresponding to Input zO

% For Ploe-zero feedback gains design we propose a Plant with Certain Design
Requirements
% Characteristic Eqn With Feedback Gains

% Let the closed loop characteristic equation for the required performance is
given by:

p1=-10;

page 149
Chapter 8 Control of the MIMO Model

P2=-15;
p3=-40 +
40i;
p4=-40 40i;
+
p5=-100 100i;
p6=-100 100i;

K=place(A,B, [p1 p2 p3 p4 p5 p6])

t=0:0.01:2;
x0=[5*pi/180 0 0 0 0 0];

Ac=[(A-B*K)];
Bc=[B]
Cc=[C]
Dc=[D]

% Feedback Gains with Reference Input

u1=a*ones(size(t) ) ;
u2=u1 ;
u3=u1 ;
' u2'

u=[u1 , , u3'];

Nbar=rscale(A,B,C,D,K)
Bcn1=Nbar(1)*B1;
Bcn2=Nbar(2)*B2;
Bcn3=Nbar(3)*B3;
Bcn=[Bcn1, Bcn2, Bcn3]

ty,x]=lsim(Ac,Bcn,Cc,Dc,u,t,xO) ;

The linear responses obtained by the pole placement design is shown in Fig. 8.9. The

figure shows that there is a lot of improvement needed in order to achieve a satisfactory

behavior. The chosen pole locations have ensured minimum settling time, but the steady
state values have suffered considerably. It requires extensive trial and error to design a set

of desired closed loop poles which will ensure the desired performance. Moreover, the

choices of N gain will result in zero steady state error to a single step input, but in the

above simulation we have considered all the three inputs, which affect the plant dynamics

causing the error to be nonzero. It has been observed that integral control provides very

good tracking of the reference input. In the next section we will introduce an integral

control along with the full state feedback.

page 150
Chapter 8 Control of the MIMO Model

Linear Responses Obtained by Pole Placement


2 0.1 1 1 1 i i \ 1 1 1
;

\ i i
D
a.
co
m 0.1
i i i i ! ! ! !
CC 0 0.2 0.4 0.6 O.B 1 1.2 1.4 1.6 1.B 2
E 1

o
cu 0.5 y^X
co

cn

0
/ i i !!!!!!
0 0.2 0.4 0.6 O.B 1 1.2 1.4 1.6 1.B 2
i i i i i i i i

CN '

^2 r

1
!"j ...

CL.

i i i 1 i i i ! !
a 1.6 1.B 2
0 0.2 0.4 0.6 0.8 1 1.2 1.4
Time (zee)

Fig. 8.8 Linear Responses of the Pole Placement Design

8.3.4 Integral Control with Full State Feedback

From the basic nature of the dynamics of the landing gear system, if we can introduce a

the displacements of the centers of the main wheel and the nose
state which will relate

we can achieve better control on the behavior of the system. Let us introduce an
wheel,

integral state corresponding to:

X!(t)
=
Zd(t)
=
ZX(t)
-

Z2(t) (8.32)

the original system that of 7th-order. The original system now can
This new state makes

which is done as follows:


be augmented to include this additional state,

x A 0 x B
+
0
u (8.33)
C[2, :
} -

C[3, : ] 0 xi
XI

page 151
Chapter 8 Control of the MIMO Model

where C[2, :] and C[3, :


] are the 2nd and 3rd rows of the C matrix corresponding to
the states zx and z2. The feedback law becomes:

x
u=
-[Kx K2 ...
K6 Ki] (8.34)
Xl

In the pole placement design, it becomes necessary to design an additional closed loop
pole corresponding to the integral state and as discussed above, a satisfactory response

can only be achieved after considerable trial and error. In the next section, we will

describe a more powerful approach called the linear quadratic regulator (LQR) method in

designing the feedback gains.

8.3.5 Linear Quadratic Regulator (LQR) Method

In this method, the feedback gains Ki are determined from the minimization of a cost

objective function J. For the state space model described by the Eqn. 8.20, the LQR

method is based on the minimization of the cost function J defined as:

/oo

J=
(xT(t)Qx(t) + uT(t)Ru(t))dt (8.35)
Jt=o

where Q is an n x n (n is the number of states) symmetric matrix and R is an m x m (m

is the number of control inputs) symmetric matrix. Usually Q is chosen as a diagonal

matrix with non-negative diagonal entries and R is also a diagonal matrix with non-

negative diagonal entries. For optimum performance, we will choose the state weighting

for the original system without the integral state is:


matrix Q =
CTC, which

0 3.8462e-020 -1.5385e-019 0
1 OOOOe+OOO -2.2316e-015

1. OOOOe+OOO 5.4729e-016 6.9792e-021 -2.7917e-020 0


0
5.4729e-016 1 -3. 5628e-020 1.4259e-019 0
-2 2316e-015 .0000e+000

6.9792e-021 2.7974e-039 -1.1192e-038 0


3 8462e-020 -3.5628e-020

1.4259e-019 -1.1192e-038 4.4781e-038 0


-1 5385e-019 -2.7917e-020

0 0 0 0 0
0

The diagonal entries Qxx, Q22, and Q33 are replaced by the scalar weighting factors p, q,

and which are manipulated to determine the optimum weighting factors for the states 9,
r,

page 152
Chapter 8 Control of the MIMO Model

zx, and z2. The matrix R is chosen as a symmetric diagonal matrix with size equal to the

number of columns in the B matrix. A value of 0.1 for the diagonal entries of the R

matrix has been found to be optimal after trial and error to represent the weighting factors

for the control inputs. Therefore the R matrix has the following form:

0.1 0 0
R = 0 0.1 0 (8.36)
0 0 0.1

Once we have defined the Q and R matrices, we can make use of Matlab's Iqr function to

find out the feedback gains. The code using LQR design to evaluate the feedback gains

and running a linear simulation is given below:

% The Following Codes Calculate the LQR Based Controller for the SS Model

% The State Space Model of the MIMO Linear Model is Given Below:

A=[4.8193e-021, 0, 0, 1, 0, 0;
0, 0, 0, 0, 1, 0;
0, 0, 0, 0, 0, 1;
-140.16, -436.42, 434.4, -0.45926, -1.5259, 1.5247;
-85.814, -951.58, 277.18, -0.29662, -1.3979, 0.97637;
32077, 1.0614e+005, -3.6422e+005, 112.49, 373.5, -373.5];

B=[0, 0, 0;
0, 0, 0;
0, 0, 0;
2.0182, 5.9853e-006, 1.2735e-013;
674.4, -0.998, 7.797e-014;
0, 0, 2.5806e+005];

C=[1 0, -2.2337e-015, 3.8462e-020, -1 0;


.5385e-019,

o', 1, 5.4729e-016, 6.9792e-021, -2.7917e-020, 0;


2.0675e-018, 0, 1, -3.5628e-020, 1.4259e-019, 0];

D=[0, 0, -9.0859e-016;
0, 0, 1.6262e-016;
0, 0, 0];

p=50; % The weighting factor for theta

q=100; % The weighting factor for z1

r=50; % The weighting factor for z2

to Input zO
B1=B(: ,D
% First Column of B matrix Corresponding
to Input zO
B2=B(: ,2)
% Second Column of B matrix Corresponding
Corresponding to Input zO
B3=B(: ,3)
% Third Column of B matrix

C matrix Corresponding to Input zO


C1=C(1 ,:)
% First Row of
Column of D matrix Corresponding to Input zO
D1=D(: ,1)
% First

'
Q=C *C J

0 -2.2316e-015
3.8462e-020 -1.5385e-019 o;
Q =[

page 153
Chapter 8 Control of the MIMO Model

o q 5 4729e-016 6.9792e-021 7917e-020


-2 o;
-2.2316e-015 5.4729e-016 r -3.5628e-020 1 4259e-019 0;
3.8462e-020 6.9792e-021 3 5628e-020 2.7974e-039 1192e-038
-1
o;
-1.5385e-019 -2.7917e-020 1 4259e-019 -1.1192e-038 4 4781 e -038 o;
0 0 0 0 0 0];

R=diag([0.1, 0.1, 0.1]);

K=lqr(A,B,Q,R)

K =

-4.2091e+001 2.2268e+002 5.7581e+000 -4.5014e+000 8.3617e-001 -2.3476e-005

1.2725e-001 -3.3028e-001 -1 .8444e-002 1.3569e-002 -1.2780e-003 8.7935e-008


4.6660e+001 1 .0097e+000 2.1561e+002 4.5871e+000 -2.271
Oe- 002 4.0111e-002

Ac=[(A-B*K)];
Bc=[B]
Cc=[C]
Dc=[D]

t=0:0.01:2;
x0=[5*pi/180, zeros(1,5)]; % Initial conditions

u1=a*ones(size(t) ); % Step Input


u2=u1 ;

u3=u1 ;
u=[u1', u2', u3'];

Nbar=rscale(A,B,C,D,K) % Calling user defined function rscale

Bcn1=Nbar(1)*B1;
Bcn2=Nbar(2)*B2;
Bcn3=Nbar(3)*B3;
Bcn=[Bcn1, Bcn2, Bcn3]
[ y , x ] =lsim (Ac , Ben , Cc , De , u , t , xO) ;

The results of the linear simulation using LQR design is shown in Fig. 8.9.

page 154
Chapter 8 Control of the MIMO Model

Linear Responses Obtained by LQR design


E D.1

0)
D -

?
Q-
0)
CD 0.1
cr 0 0.2 0.4 D.B O.B 1 1.2 1.4 1.B 1.B 2
D.D2

CD
ID
a
o
Q-
V>
CD
0.02
cr
0 D.2 0.4 O.B O.B 1 1.2 1.4 1.B 1.B 2
0.D5 i 1 1
t-
1 1 1 1 1 r

rsi

CD
en
c
?
CL.
w
CD J L i i I L
-0.D5
cr 1.2 1.4 1.B 1.B 2
D 0.2 0.4 O.B O.B 1
Time (sec)

Fig. 8.9 Linear Responses of the LQR Design

It can be seen that the LQR design has clearly improved the responses in terms of the

settling time and the steady state errors. It also gives us the flexibility to randomly choose
to manipulating the roots of the closed
the scalar weighting factors p, q, and r as opposed

done in the case of pole placement design. In order to understand the


loop system as was

nature of the cost function J, a Simulink model is created as shown:

page 155
Chapter 8 Control of the MIMO Model

?J_
5cope4 Matrix
Gain

G-

wgran*.l>

Jart(0

LQR Integral

lntegrarid(1)

Jeosttt)
Integrator

Fig. 8.10 Simulink Diagram to Implement LQR design

the function is given in Fig. 8.11. The cost function is an


A plot of the integrand and cost

function and attains state values for each control input. In our case, the
increasing steady

found to be 0.0078067, 0.0077585, and 0.007766. If we increase


steady state values are

(diagonal entries of the Q matrix) it can be seen that the


the weighting factors p, q, or r

state decrease, but there may be couplings with other


transients in the corresponding

this. At the same time, increasing the weighting factors


states, which maybe affected by
function. Similarly, by increasing the diagonal entries of
will result in an increased cost

particular control input decreases.


the R matrix, the gain corresponding to that

page 156
Chapter 8 Control of the MIMO Model

Evaluation of the Cost Function, J


?.16 1 i i

0.14
'

0.12 l ; ,

i i
->_ 0.1
l :
c

1 Integr; ind
H 0.08
CD

1 0.06

0.04 h

Cosf Function
0.02
V
*,
_ yf i i
i -i i i i

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1


Time (sec)

Fig. 8.1 1 Cost Function and the Integrand

8.3.6 LQR Design with Integral Control

Let us introduce the integral state defined earlier to our LQR design. Since the

introduction of the integral state renders the original system that of 7th-order, we will

need to augment the Q matrix to include a weighting factor (say p) to represent this state.

Therefore, our new Q matrix has the following form:

q= 0 3.8462e-020 -1.5385e-019 0 0;
[ p -2.2316e-015

0 5.4729e-016 6.9792e-021 -2.7917e-020 0 0;


q
-2.2316e-015 5.4729e-016 r -3.5628e-020 1.4259e-019 0 0;
3.8462e-020 6.9792e-021 -3.5628e-020 2.7974e-039 -1 .1192e-038 0 0;
-2.7917e-020 1 .4259e-019 -1 . 1 192e-038 4.4781e-038 0 0;
-1.5385e-019

0 0 0 0 0 0;
0
0 0 0 0 0 rho];
0

The A matrix is modified to include the new state as:

Ae = (8.37)
C[2, :
} -

C[3, :
}

page 157
Chapter 8 Control of the MIMO Model

The integral state is not considered in calculating N because the 7th row of the A matrix

with the integral state added, which is nothing but C[2, :


] C[3, : ], is linearly
dependent on the rows of the original A matrix thereby making it a singular matrix. The

rows of the C matrix are modified to include the weighting factors of the states before

passing it into the r scale function. The codes to calculate the feedback gain using LQR

design with integral control are shown below:

% To Find the State Space Model for the new system

Ae=[A, zeros(6,1);C(2, :)-C(3, :), 0]; % Augmented matrix with integral state

Be=[B; zeros(1,3)]; % Augmented matrix with integral state

rho=5000; % The weighting factor for the integral state


p=5; % The weighting factor for theta
q=30; % The weighting factor for z1

r=10; % The weighting factor for z2

Q= 0 2.2316e-015 3.8462e-020 -1.5385e-019 0 0


t P
0 5.4729e-016 6.9792e-021 -2.7917e-020 0 0
q
2.2316e-015 5.4729e-016 r -3.5628e-020 1.4259e-019 0 0

3.8462e-020 6.9792e-021 -3.5628e-020 2.7974e-039 -1 . 1 192e-038 0 0

1.5385e-019 -2.7917e-020 1 .4259e-019


-

1 . 1 192e-038 4.4781e-038 0 0

0 0 0 0 0 0 0

0 0 0 0 0 0 rho] ;

R=diag([0.1, 0.1, 0.1]);

K=lqr(Ae,Be,Q,R);

K =

1.7486e+001 -1.0757e-001 -1.0812e-001 2.2681e-001 -2.1055e-005


-2.5393e+000

8.9376e+001
3.8212e-005 2.4266e-004 -3.3636e-004 3.3184e-008
3.8584e-003 -2.5819e-002

'

8.6661e+000 1.7473e-001 -8.5795e-003 6.9941e-003


6.0234B+000 -1.65219+000

2.0497e+002

t=0:0.01:2;
zeros(1,6)]; % Initial Conditions
x0=[5*pi/180,
% Step Input
u1=0.0025*ones(size(t) ) ;
u2=u1;
u3=u1 ;
'
u=[u1 , u2\ u3'];

% Input Scaling

of the original six states


Kn=K(:,(1:6)); % Only the gains

factors
% The C matrix is modified to include the weighting

0;
Ce=[p, 0, -2.2337e-015, 3.8462e-020,
-1 .5385e-019,

5.4729e-016, 6.9792e-021, -2.7917e-020, 0;


0, q,
1.4259e-019, 0] ;
2.0675e-018, 0, r, -3.5628e-020,

page 158
Chapter 8 Control of the MIMO Model

Nbar=rscale (A, B, Ce , D, Kn )
Bcn1=Nbar(1)*B1;
Bcn2=Nbar(2)*B2;
Bcn3=Nbar(3)*B3;
Bcn=[Bcn1, Bcn2, Bcn3];

% Form Controller Matrices

Ac=[Ae-Be*K];
Bc=[Bcn; zeros(1,3)];
Cc=[C zeros(3,1);zeros(4,7)];
Dc=[D; zeros(4,3)];

[y,x]=lsim(Ac,Bc,Cc,Dc,u,t,xO);

The result of the simulation is shown in Fig. 8.10.

Linear Responses Obtained by LQR design

? ,--30.2 0.8 1.2 1.4 1.B 1.8


x 10
E 10

_.

j 1 ( ( T n ( 1 r
Jl i i i i i i i i i

CD
| | i
i
i
i
i
i
i
i
i
i
i i i i

I i i i i
en \ i ^y
~^r^-~_^_ i i i i i i i

a
a.
en
CD
cr
0.2 0.4 O.B O.B 1.2 1.4 1.6 1.B
E 0.02
CM
N

i i
',
T i i i i i i >
r i i i i i i i i i
CD J i i i i i i i i i
i i i i i i i i
CO i

-0 02 J , , T t ! 1 r

a
Q-
01
cu n
CT
u 04
0 0.2 0.4 O.B D.B 1 1.2 1.4 1.B 1.B

Fig. 8.12 Linear Simulation Results of the LQR Integral Control

page 159
Chapter 8 Control of the MIMO Model

The effect of the integral state is not quite apparent in the case of linear simulation,
because we have only considered step inputs. But it has been included to make the

controller robust under varying input conditions. As a next step towards improving our

controller we will design an estimator which will provide an estimate of the states of the

system in case we do not have all the states available for feedback.

8.3.7 Estimator Based LQR Design with Integral Control

In most cases, not all the state variables are measured. The cost of the required sensors

may be very high, or it may be physically impossible to measure all the state variables. In

this case, an estimate of the actual state variables x(t) can be obtained by a few

measurements. If we represent the estimate of the actual state as x(t), the control law

without the reference input becomes:

u= -Kx.
(8.38)

The plant equation with feedback is now:

Ax-
x =
BKx, (8.39)

which can be rewritten in terms of the state error x = x x (actual state estimated

state) as:

Ax-
x = BK(x -

x) (8.40)
= (A- BK)x + BKx

In order to find the error state equation, we build the plant dynamics in terms of the

estimate of the actual states as:

x = Ax+Bu (8.41)

back the difference between the measured and the estimated


If we consider feeding
outputs and correcting the model continuously with this error signal, the above equation

will be modified as:

page 160
Chapter 8 Control of the MIMO Model

x =
Ax +Bu + L(y-Cx), (8.42)

where L is the proportional gain defined as:

L=[lx,l2,...,lnf (8.43)

and is chosen to achieve


satisfactory error characteristics. The dynamics of the error

system can be obtained by subtracting Eqn. 8.42 from Eqn. 8.20 as:

x=(A-
LC)x (8.44)

The overall system dynamics in state form the estimator are obtained
with
by combining
Eqn. 8.40 and Eqn 8.44 to get:

X A-BK BK X

X 0 A-LC X
(8.45)

The characteristic equation of this closed loop system is:

sI-A + BK -BK
det (8.46)
0 sI-A + LC

This particular form of the above matrix allows independent design of the control law and

the estimator. The combined control law and the estimator is called the compensator. The

block diagram representation of feedback control with the estimator is shown in Fig. 8.13.

page 161
Chapter 8 Control of the MIMO Model

Plant

u(t) x=Ax+Bu x(t) Sensor y(t)


-o
y=Cx+Du C

u(t)

Estimator
Control Law M) x=Ax+Bu
u--Kx
+L(y-C)

Compensator

Fig. 8.13 Feedback Control with the Estimator

If we include the integral control, Eqn. 8. 45 will be modified as follows:

B~

X A 0 X
[K Ki] -

K
XI =
C[2, :
] -

C[3, :
] 0 0 XI (8.47)
0 -LC X
X .

Matlab codes are created to calculate the proportional feedback gain L. In


The following
locations are chosen such that they are 4-10 times
order to design an estimator, the pole

faster than the slowest pole of the closed loop system [14].

% To Design an Estimator

% Eigen values of the closed loop system


p=eig(Ac)

=
P

-
1 . 0752e+003 + 1 .
2042e+003i
-
1 0752&+003
.
-
1 .
2042e+003i

-7.3245e+001
+7.9951e+001i

-7.9951e+001i
-7.3245e+001

-2.2444e+001

-6.8527e+000 +1 .0572e+001i
% Slowest Poles
-6.8527&+000
-1.0572e+001i % Slowest Poles

% Chosen estimator pole locations


P=[-20 -21 -22 -23 -24 -25];

L=place(A',C,,P)'

Ace=[Ae-Be*K Be*K( : ,
(1 :6) ) ;zeros(6,7) (A-L C)];

page 162
Chapter 8 Control of the MIMO Model

Bce=[Bcn; zeros(size(Be) )];


Cce=[C zeros(3,7) ] ;
Dce=[D];
[ y , x ] =lsim (Ace , Bee , Cce , Dee , u , t , xO );

The estimator design is verified in Simulink (Fig. 8.14) and a comparison of the actual

states and the estimated states is shown in Fig. 8.15. The plot of the comparison shows

that the estimated and the actual states have a very good match.

LQR, integral Feedback


of the Linearized 3 DOF Model
with Estimator
?
ScopeS

tf= Ax*Bu
Nbar
-> y=
Cx*Du

Step Stale-Space
Gain
> z2

th_h

z1_h

z2_h

thdji
Subsystem

rldji

-
z2d_h
To WorispaceS

Estimated States

Feedback Loop

page 163
Chapter 8 Control of the MIMO Model

Estimator Design

n
cop2

J}~
Integrator
L<j Gx-yjD

Ujtnx
Gjir\2

Matrix
0ain3

Fig. 8.14 Simulink Block Diagram to Design an Estimator

Actual State
Estimated State
0.15
w
co
~^-
U.1
-c;

r-

03
tf)
nns L j [
C
n
Q-
rr) 0 \ [ j
q: ^/T i
-0.05

1 1.5 1.5

S 0.2 -i i -j

"- 0
0)
w

1-0.2 fi
cr
-0.4

1 1.5 I 0.5 1.5

0.02 20

I 0 r- 0
N

S -0.02
":--2o
CO

-0.04
1-40
a)

cc
60
-0.06

0.5 1 1.5 0.5 1.5

Fig. 8.15 Comparison of the Actual and the Estimated States

page 164
Chapter 8 Control of the MIMO Model

So far we have discussed the control of the linearized 3-dof model using the pole

placement method and the LQR method. It is quite obvious that the LQR method is faster
in terms deciding the feedback
of gains necessary for optimum performance. We have
also improved the controller robustness by introducing the reference input and integral
feedback. Now we are ready to apply the designed controller to the original nonlinear 3-

dof model with all the inherent dynamics present in the original model.

8.4 Control of the Nonlinear 3-dof Model

8.4.1 Simulation with Step Input

We will first analyze the performance of the controller on the nonlinear model with step
inputs in place of the sinusoidal inputs [22]. Consider the case when the airplane is

taxiing on the runway after the initial stage of landing. Let us simulate this case by giving
two step inputs to the ODE's of zx, and z2. Even though, the ODE of 9 has a second

derivative of the input zq, given at zx, the effect of a step input is negligible on the

equation of 9. This apparently means that we need to calculate three input scaling factors

(N) and three feedback loops, one each for each input. One important thing to be kept in

mind is that due to the inherent dynamics of the airplane model and the approximations

taken while deriving the equations of motion, the steady state conditions of the states are

different from zero. The following Simulink diagram (Fig. 8.16) shows the

implementation of the LQR controller along with integral feedback for a step input
magnitude of 0.0025 m and the responses are compared with those of the linear model as

shown in Fig. 8.5.

page 165
Chapter 8 Control of the MIMO Model

LQR Design with Integral Feedback


on the nonlinear 3-dof model

Integrator!

Fig. 8.16 Simulink Diagram of LQR Controller with Integral Feedback

page 166
Chapter 8 Control of the MIMO Model

Nonlinear Response
Linear Response
0.15 1 1 1 1 1

JS 0.1 r
-
-

0.05 ^y
-
-
-r^--- ,-

, --, ,

o
V /.-k. ; ; ; ;
Q- 'i ' ' ' ' '
/
C/> '-'
<X>
CC -0.05
i i i i i 1
0.5 1.5 2.5 3.5 4.5

0.02 1 1 1 1 1 1 1 1 1
1 1 1 1 t 1 1 1 1

0.01

V I i i i i i J i
-0.01
i
0.5 1.5 2.5 3.5 4.5

0.01

-0.01

-0.02
V-

CC
_L

-0.03

0.5 1.5 2 2.5 3 3.5 4.5


Time (sec)

Fig. 8.17 Controller Performance on Nonlinear and Linear Models

responses of the linear and nonlinear 3-dof model shows that the
A comparison of the

both is very similar in terms of the transient as well as the steady


behavior of models

that the controller developed based on the linearized version


state responses. This shows

3-dof model works well on the nonlinear system. Due to the inherent
of the nonlinear

model, their steady state values are different from


nonlinearities present in the nonlinear

have met the design requirement of 1 % settling


zero. It can be seen that the responses

But the have gone beyond the 25 % design benchmark


time of less than 2 sec. overshoots

trade off between the overshoot and the settling time. The weighting
as there is always a

with a reasonable overshoot limit. The above


factors maybe manipulated to come up

simulation was carried with the following weighting matrices:

page 167
Chapter 8 Control of the MIMO Model

5, .OOOOe+OOO 0 -2.,2316e-015 3 ,8462e-020 -1 ,5385e-019 0 0


0 3. ,0000e+001 5 .4729e-016 6, ,9792e-021 -2 ,7917e-020 0 0
-2 ,2316e-015 5, ,4729e-016 1 .0000e+001 -3, 5628e-020 1 .4259e-019 0 0
3 ,8462e-020 6 ,9792e-021 -3 ,5628e-020 2. ,7974e-039 -1 ,1192e-038 0 0
-1 .5385e-019 -2.,7917e-020 1 .4259e-019 -1 .
,1192e-038 4 4781 e
. -038 0 0
0 0 0 0 0 0 0
0 0 0 0 0 0 7.0e+003

1.0000e-001 0
0 1.0000e+001 0
0 0 1 .7000e+001

By increasing the weighting factor corresponding to a particular state, the transients in the
response due to that particular state can be reduced, but it may affect other states which

are coupled with the original state. Higher values are chosen for R22 and R33 compared to

Rxx to minimize the effects of the control inputs due to the inputs at zx and z2. The above

simulation with step input is only shown as an illustration of how the controller works

with the nonlinear system. In order to achieve the final objectives, the nonlinear system

has to be simulated with the assumed road profile sinusoidal inputs. Therefore, the issues

of overshoot and settling time will be dealt with in greater detail when we simulate the

final system. In order to apply the sinusoidal inputs, a more effective method is described

in the following section.

8.4.2 Robust Control Based on Error Space Approach

It should be noted that the input scaling factors (N) discussed in the previous section can

only be used with step inputs to obtain zero steady state error. For sinusoidal inputs, a

good reference input tracking can be achieved by introducing an internal model of the

reference input in the compensator. The entire problem is formulated in an error space to

obtain zero state errors to non decaying reference inputs. To begin with, the
steady
method is described in general first and then it is applied to the particular problem at

hand.

Suppose we have the system state equations with a disturbance, w, and a coefficient

matrix for disturbance, Bx, represented by:

x(t)
=
Ax(t) + Bu(t) + Bxw (8.48)
y(t)
=
Cx(t) + Du(t),

page 168
Chapter 8 Control of the MIMO Model

and a reference input r(t), which is known to satisfy a specific differential equation. The
initial conditions on the equations
generating the input may not be known to us. The
objective is to design a controller for this system so that the have
closed loop system will

the ability to track input command signals and to reject the disturbances without steady
state error. Suppose the reference input satisfies the relation:

r + axr + a2r = 0 (8.49)

and the disturbance satisfies exactly the same equation:

iv + axw + a2w =
0 (8.50)

The tracking error of the system is defined as:

e =
y-r.
(8.51)

The specific objective of tracking r and rejecting w can be described as regulation of

error, which means the error tends to zero as time gets large. The control must also be

robust enough that the regulation of error to zero occurs even in the presence of small

disturbances in the plant parameters. In practice we can not have a perfect model of the

plant and values for the parameters.

Substituting for r in Eqn. 8.49 from Eqn. 8.51, we get:

e + axe + a2e =
y + ax y + a2y (8.52)
= Cx + axCx + a2Cx + Du + axDii + a2Du

We now replace the plant state vector with the error space state defined by:

fa x + axx + a2x.
(8.53)

Similarly, we replace the control input with the control in error space, defined as:

a fa u + ot\u + a2u.
(8.54)

With these definitions we can replace Eqn. 8.52 with:

page 169
Chapter 8 Control of the MIMO Model

e + axe + a2e =
C+ D/j,. (8.55)

The state equation for is given by:

=
x+ axx + a2x =
A + Bp. (8.56)

Notice that the disturbance as well as the reference cancels from Eqn. 8.56. Equations
8.55 and 8.56 now describe the overall system in an error space. In standard state variable

form the equations can be represented by:

z =
Aez + Bep, (8.57)

rp
r
where z =
[ e e ] and

'B~

A 0 0
Ae = C Oi2 -ax , Be = D (8.58)
0 1 0 0

Now a full state feedback can be designed for the error system (A, B) if it is controllable.

If the plant (A, B) is controllable and does not have a zero at any of the roots of the

reference signal characteristic equation:

s2

ar(s)
=
+ axs + a2, (8.59)

then the error system is controllable. Assuming these conditions hold, there exists a

control law of the form:

H=
-[Kq Kx K2\ Kz. (8.60)

Where Kq are the feedback gains for the plant states and Kx and K2 are the feedback

gains for error signals. In the next section, we will implement the error space approach to

our nonlinear 3-dof model with sinusoidal reference inputs.

page 170
Chapter 8 Control of the MIMO Model

8.4.3 Error Space Approach Applied to Nonlinear Model with Sinusoidal Inputs

Since the error space approach is based on


expressing the input signals in terms of

differential equations, we need to first come up with the state space model representing
the input signals. There are three sinusoidal input signals to our nonlinear model:

zq =
a sin(u>
t) (8.61)
u2

= a
zq sin(u>
t)
-

f2-7T
zh = a -
sm<
vo-t+ (PD -

QF)sin9 + QP cos9
I

(QF cos9 + QP sin9


-

PD cos9 + zx-z2
ED\ 1
V ED ))}
To get an idea of the amplitude and frequency of the above signals, a plot of the input

signals is shown in Fig. 8.5 in which the variable zOd represents zq.

x
ID-

Input Signals zO, zOd, and zh

V \.i.J Lj. J l_L.i...J-l- J l-L-t.-i-L


\.'--f VJ-f V J--h

0.1 0.2 0.3 0.4 0.5 O.B 0.7 O.B 0.9 1


Time (sec)

Fig. 8.18 Input Signals Generated for 1 sec

page 171
Chapter 8 Control of the MIMO Model

By looking at the three signals, even though the amplitudes are different, their frequencies
are nearly the same. The periods of the signals are 0.1 sec and the frequencies are 10 Hz.
The amplitudes of the three signals z0, z0, and zH are 0.0025 m, 10 m/s2, and 0.0025 m.
This implies that all the three input signals can be written as second order differential
equations satisfying the relation described by Eqn. 8.49. Although, zH is represented by a
complex mathematical form, we can approximate it by a simple sinusoidal signal for our

analysis purpose. From the lagging nature of zH, we can approximately represent this

signal as:

zh ~
-

a cos(ua -t + r), (8.62)

where r is the phase lag and the frequency ua is approximately equal to u.

Let us denote the three signals zq, zq, and zh by rx, r2, and r3. We first find the

derivatives of these signals as represented in the following Table 8.1 :

oj2

rx =
a sin(ui
t) ?"2 =
a sin(cu t) r3 =
a cos(ua -t + r)
fx= to3

f3 sin(uja + t)
a-uja-
U a oo cos(u t) r2 =
a cos(u t) =

cos(ua + r)
co2

ri rx a sin(u t) r2 = a a/ sin(uj t) r3 = a
u\
= = r3=
rx u2.rx r2 -r2 -u2a-r3

Table 8.1 Derivatives of the Input Signals

Now we form the state space system to represent this reference input signals as:

rx = x r,> *
^
-

<X'7"-j
*~
w
rx = x.n>
(8.63)
x r2> r2 x rs>
r2 r2 Xr2 Xr^ ,

= x.^6'
r3 = x ^3' Xr3 XT6, r3

which can be represented in matrix form as:

page 172
Chapter 8 Control of the MIMO Model

0"

0 0 0 1 0
xri
x
"2
0 0 0 0 1 0 xt2
x 0 0 0 0 0 1
~-

/l<i//f xr3
(8.64)
- ~"-
%JL"p
X 0 0 0 0 0 Xt\
X
^5 0 0 0 0 0 xTh
_Xr6
0 0 0 0 0_ _xTft
_

10 0 0 0 0 ^r2
X
Tz
/ Ci -p *b j*
0 10 0 0 0
X Ti
0 0 10 0 0
X
Ti

xr&

In input-output form, the reference model in Eqn. 8.64 is given by:

T\ 0 0 rx
r2 =
0 0 7*2 (8.65)
rz 0 0 ?"3

Since there are three outputs and three inputs, the tracking error has nine components as

shown in the following matrix:

9-zq eZlzo
= ZX
Zq e-zizo
=
z2 zq
e =
e9z0 9-zq eZl20
=
zx zq CziZo =
Z2 zq (8.66)
e-ezH
= 8 ~

zh ez,ZH
=
zx zh eZ2Z =
z2
-

zH

Taking the time derivative of the generic form of the tracking error twice yields:

e =
y r (8.67)
=
Cx + Du r,

Utilizing Eqns. 8.52 through 8.57, we can rewrite the overall system in matrix form as

follows:

page 173
Chapter 8 Control of the MIMO Model

l6x6 06x6
0 0 0 0 0
^6x1 C3xt 0 0 0 0 0 6x1 #6x3
<23xl 0 0 0 0 0 a
e3xl + D /i.(8.68)
<33xl 1 0 0 0 0 0 e3xi 0
03 x6 0 1 0 0 0 0
0 0 1 0 0 0

The overall system is now 12th-order. Assuming that the system is controllable, we can

design a control law of the form:

P =
-[#0(3x6) #1(3x3) #2(3x3)] = -Kz.
(8.69)

The error space approach in its original form as described above, can be implemented for

simple systems to track reference inputs and reject the disturbance signals. But it becomes

very difficult to apply the exact form to the particular problem at hand, because there are

three inputs. Moreover since the above model has 12 states, the overall system becomes

more coupled and thereby difficult to control. In order to avoid these difficulties, a

simplified form of the error space approach is used on the nonlinear 3-dof model. In this

model, integral feedbacks of the error signals are used. This can be done by considering

only the following part of Eqn. 8.68:

^6x6 06x3
0 0 66x1
66x1 +
-56x3

(8.70)
C3x6 0 0 e3xi D
e-3xi
0 0 cuo

and the control law becomes:

H
= -

[#0(3x6) #1(3x3)] = -Kz.


(8.71)

The overall system is now 9th-order and easier to control. The following codes in Matlab

show the formulation of the new system and calculation of K. A linear simulation is

feedback gains. The responses are compared (Fig. 8.19)


performed using the calculated

page 174
Chapter 8 Control of the MIMO Model

with that of the LQR design with one state integral feedback discussed in the previous

section.

% Codes to Evaluate the Feedback Gains for the Error State Space Design

% State Space Model in Generic Form for I.C on Theta=0.3048 rad/sec

A=[4.8193e-021, 0, 0, 1, 0, 0;
0, 0, 0, 0, 1, 0;
0, 0, 0, 0, 0, 1;
-140.16, -436.42, 434.4, -0.45926, -1.5259, 1.5247;
-85.814, -951.58, 277.18, -0.29662, -1.3979, 0.97637;
32077, 1.0614e+005, -3.6422e+005, 112.49, 373.5, -373.5];

B=[0, 0, 0;
0, 0, 0;
0, 0, 0;
2.0182, 5.9853e-006, 1.2735e-013;
674.4, -0.998, 7.797e-014;
0, 0, 2.5806e+005];

C=[1, 0, -2.2337e-015, 3.8462e-020, -1 0;


.5385e-019,

0, 1, 5.4729e-016, 6.9792e-021, -2.7917e-020, 0;


2.0675e-018, 0, 1, -3.5628e-020, 1.4259e-019, 0];

D=[0, 0, -9.0859e-016;
0, 0, 1.6262e-016;
0, 0, 0];

% To Find the State Space Model for the Eorro Space system

om=diag(-omega"2*[1 ,1, 1]);


id=diag([l, 1, 1]);
Ae=[A, zeros(6,3);C, om]
Be=[B; D] ;
De=[D];

rho1=1; % The weighting factor for zOd (edot)


rho2=5; % The weighting factor for zO (edot)
rho3=5; % The weighting factor for zh (edot)

p=5-
% The weighting factor for theta

q=20; % The weighting factor for z1

r=50; % The weighting factor for z2

3.8462e-020 -1 0 0 0 0;
n= r n o -2.2316e-015
.5385e-019

1
5.4729e-016 6.9792e-021 -2.7917e-020 0 0 0 0;
Q q
r -3.5628e-020 1.4259e-019 0 0 0 0;
-2 2316e-015 5.4729e-016
2.7974e-039 -1 . 1192e-038 0 0 0 0;
3 8462e^020 6.9792e-021 -3.5628e-020

4.4781e-038 0 0 0 0;
-l'5385e-019
-2.7917e-020 1 .4259e-019 -1 .1192e-038

0 0 0000 0;
0 o
0 0 0 0 rhol 0 0;
0 0
0 0 0 0 0 rho2 0;
0 0
0 0 0 0 0 0 rho3] ;
0 0

R=diag([0.0l, 0.01, 0.01]);

page 175
Chapter 8 Control of the MIMO Model

K=lqr(Ae,Be,Q,R);

K =

Columns 1 through 7

-1.0148e+001 4.5822e+001 1.4165e+000 -1.2592e+000 3.7775e-001 -1.8967e-005

1.3653e-009
1.8609e-002 -6.8604e-002 -2.6442e-003 2. 2651 e- 003 -5.6580e-004 3.7726e-008
3.9581e-013
8.1119e+000 -8.0162e-001 6.8176e+001 8.3282e-001 -9.7500e-003 2.1796e-002
2.3488e-008

Columns 8 through 9

2.5697e-006 4.2523e-007
-3.8004e-009 -8.0836e-010

7.8849e-008 2.8385e-004

K0=K(:,(1:6));

K1_zO=K(1,7:9);
K1_th_z0=K1_z0(1);
K1_z1_z0=K1_z0(2);
K1_z2_z0=K1_z0(3);

K1_z0d=K(2,7:9);
K1_th_z0d=K1_z0d(1);
K1_z1_z0d=K1_z0d(2);
K1_z2_z0d=K1_z0d(3);

K1_zh=K(3,7:9);
K1_th_zh=K1_zh(1);
K1_z1_zh=K1_zh(2);
K1_z2_zh=K1_zh(3);

t=0:0. 001:2;
x0=[5*pi/180, zeros(1,8)];
u1=a*sin(omega*t) ;
u2=-a*omega*2*sin(omega*t) ;
u3=-a*cos(omega*t) ;
u=[u1', u2', u3'];

% Form Controller Matrices

Ac=[Ae-Be*K];
Cc=[C, zeros(3,3)];
Dc=[D];

[ y , x ] =lsim (Ac , Be , Cc , Dc , u , t , xOi) ;

page 176
Chapter 8 Control of the MIMO Model

Linear Responses Obtained by LQR with Error Space Approach


S 0.1
LQR with Error Space

cd
CO
c
o
0-
u LQR with one state Integral FB

Q.
CO
0.1
i i i i i i
a 0 0.5 1.5 2 2.5 3.5 4.5
? 0.02

../a*

co
c
o
Q.
CO
a>
0.02 J L J L J L
cr
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
0.01
CM
N
iXJ
xzTT
0)
CO
c O.01 -

o
a.
co
CD _L
I _L

-0.02
CC 4.5
"0 0.5 1 1.5 2.5 3.5

Fig. 8.19 LQR Error Space vs. LQR Integral One State Feedback

The LQR error space approach has definitely improved the responses in terms of the

overshoot in case of 9, but has paid price in terms of the overshoot of zx. The main reason

why we are not able to improve the three responses simultaneously is because the

dynamic equations of motion are coupled in the three coordinates of the system. If we try

to reduce the overshoot in one coordinate by adjusting the corresponding weighting

of other coordinates suffer. This means that it is not possible to


factor, the responses

improve the responses after certain limits in terms of overshoot and settling time with the

dynamic equations of motion. As we know, a good control system design comes


current

after considerable practice and design iterations, the model can be updated or modified to

or stiffness as per the specific practical design requirements.


include more damping

to the nonlinear model, we need to design feedback loops


In order to apply the controller

in to the original state feedback loops. There are three rows


for the error signals addition

page 177
Chapter 8 Control of the MIMO Model

in the K matrix, each one of which represents the control input in the order =
uTl z0;

uT2
=
z0, and uTz =
zH. The first six columns of the K matrix are the gains for the 6

states of the plant model. The first six columns of the K matrix are multiplied with the
six plant state signals to obtain the three feedback control input signals from the six states

of the system. The feedback gain values to the error signals are the entries
corresponding
from 7th to 9th-columns of the K matrix. The specific K value is multiplied with the

corresponding error signal to obtain the feedback signal for that corresponding reference

input signal. The following matrix equations show the multiplications of the feedback
gains with the state the
corresponding signals, as well as error signals:

xx
x2
u
XZq
X3,
u
XZ0
=
K(:,1:6) 4
(8.72)
luxzH
X5
Xq

u,ezo eezo
= 9 -

Zq eZlzo
=
ZX Zq eZ2ZQ
=
Z2 Zq
u.ez0 K(:,7:9) eezo
= 9 -Zq eZlz0
=
ZX Zq e^
=
Z2 Zq
U,ezH ee zh =
9 zh *ZlzH
=
zx zh ez2zH
=
z2 zh

Fig. 8.20 shows the implementation of the LQR based error space approach on the

nonlinear 3-dof model and the responses are shown in Fig. 8.5. The subsystems for

multiplying the error signals with the corresponding feedback gains are shown below the

first block diagram. These subsystems are named as input scalings in the parent block

diagram. The control inputs going to the plant model are denoted by the variables u-zOd,

u-zO, and u-zh in the block diagram. These control inputs can be calculated as follows:

xx
x2 -1 T
u
20 eezo 'Z\Z0 -Z2ZQ
X3
K( 1 6)
: :
#(:,7:9)/ ee-zo
-

u = -

20 , 'ZxZq 'Z2Z0 rir(8.73)


4 Jo
u
ZH .
eez 'Zl-Zff '22-Zff .

X5

Xq

page 178
Chapter 8 Control of the MIMO Model

? 1
|

K\. zOd
U_2fld
th th
1 V
u.zDd zOdotdot
zl Z1
"I* thzl ;2 ik
u_zQ

Input Scaling zOd r a z2

te^S '
>Ki)
ft,
1 V
zO thdot
u_zh
u_zO
AJ mod el

T* thz1z2 j*/T> r zh
zldot zldot

l k
trtj ui Sealing zO z2dot ->z2dot

Plant Model
zh

u_zh

3*
thil 22

ln( ut Scaling zh

H-

3
thj! z2

-zh
J
*
1
time _,
-? time
zh mode

-3

(TV * r^j?u
vy +T
Matrix
Gain

r-

LQR with Error Space Approach


on the nonlinear 3-dof model

Integral of Error
ZO-th, 20-21, 20-22 Integral of Error
zOd-th, 200-21, z0d-z2

a> ?
*Q

QD1

r
*- -*-Zy>
? w j \ -KID
u_zOd

*
-*Q

page 179
Chapter 8 Control of the MIMO Model

Integral of Error
zh-th, zh-z1, zh-z2

Q>

Th ?

Scopel

ts^* Ki
-^>
+(Q ? -KjD

?) ? -1-J2^^>

Fig. 8.20 Implementation of LQR Error Space Approach on Nonlinear Model

Responses of th, z1 , and z2

oi 0.05
Mp=24% ,.S=0.01479

0.5 1.5 2.5

D.02
Mp=3Bc
'o
&=0TJ029~
T3 0.01 ^b

1 /...tt:

-0.01
v/
0.5 1.5 2.5
10'"

s.s=-o.ooioa
H-
0 ^d:

-5

Mp=4S %
-10

-15

D.5 1.5 2.5

Fig. 8.21 Control of the Nonlinear 3-dof Model

page 180
Chapter 8 Control of the MIMO Model

The above responses are obtained after


trying various combinations of the weighting

factors for the states of the system. A comparison of the performance of the controller on

the linear as well as the nonlinear models with the same is in


tuning parameters shown

Fig. 8.23. The figure shows a close match between the linear and the nonlinear models in
terms of the trends of the responses, but differ in steady state values. The difference in

steady states is due to the inherent dynamic properties of the nonlinear model.

Linear vs. Nonlinear Responses: LQR with Error Space Approach


=? 0.15
ro Linear Response
^ 0.1 Nonlinear Response

0.05 V : \ [ L J I
\ v-^-
i
!
i
-. :
i
:
i i
: : :
^
::::::: :
^
-0.05
I i i I I i i i
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

-. 0.02
E
0.01
j
Jf~^~~i'
a, 0 |

-0.01 -1 \jj \ \ \ \ \ \ [ j
I I I i i I I !
-0.02

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

0.01
,
CN
o B^J_i
'^=p=r
I ::::::

r
p j j j | j
-

-0.01

I I I I I i i i i
-0.02

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5


Time (sec)

Fig. 8.22 Linear vs. Nonlinear Responses

The performance of the LQR based controller can be improved further by introducing
(x4, x5, and x6) of the plant model in the Q
weighting factors for the remaining 3
states

matrix. In fact, by doing this we are actually reducing the transients due to that particular

state variable. Considerable improvement in the case of overshoots in this particular

weighting factor for only x4. The Q matrix is


example can be seen by introducing a

modified to include a weighting factor for x4 (say s) and the new responses are plotted in

Figs. 8.23, 8.24, and 8.25.

page 181
Chapter 8 Control of the MIMO Model

0=
[ 0 -2.2316e-015 3.8462e-020 -1 .5385e-019 0 0 0

0 q 5.4729e-016 6.9792e-021 -2.7917e-020 0 0 0 0


-2.2316e-015 5.4729e-016 r -3.5628e-020 1.4259e-019 0 0 0 0
3.8462e-020 6.9792e-021 -3.5628e-020 s -1.1192e-038 0 0 0 0
-1.5385e-019 -2.7917e-020 1.4259e-019 -1.1192e-038 4.4781e-038 0 0 0 0
0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 rhol 0 0
0 0 0 0 0 0 0 rho2 0
0 0 0 0 rho3];

Responses Obtained After Adding s=0.031in Q matrix

Mp=6%--r
SS=0.0l47

A
0.2 0.4 D.B O.B 1 1.2 1.4 1.B 1.B

0.02 i i i i

Mp=25.99 i i
%
RRin nn?RR
t= 0.01 i

\ ' \/ \
-1

cr
^^
-0.01
i 1 i i i i
0 0.2 0.4 O.B O.B 1 1.2 1.4 1.B 1.B

I i
i
! ! E S=-0.00l79
~
0 -i r ^^r
/ 'v^^

-5
/ i

-1D
t i
cc ^\iyip=47%
i i i i i i
-15

0.2 0.4 0.6 O.B 1.2 1.4 1.B 1.B

Fig. 8.23 Responses Obtained After Adding s =


0.031

page 182
Chapter 8 Control of the MIMO Model

S D.2

Z. 0
o

1-0.2

?
Q-
w
-0.4 SHzb
ng
D D.2 0.4 O.B O.B 1 1.2 1.4 1.B 1.B 2

"i i I 1 1 1 1 1 r
1
1 1--

\r j i 1
ce -1
1 1 1 i_
0 0.2 D.4 0.6 08 1 1.2 1.4 1.6 1.B 2

g-
10
~
E
0
a
-a
CN
~
-10

-
-20

cr .30
Li 0.2 0.4 0.6 O.B 1 1.2 1.4 1.6 1.8 2
Time (sec)

Fig. 8.24 Plot of the State Variables x, x$, and xq

0.5
1

D r t

1 I I I
1^

I 1 1 L
-0.5

? D.2 0.4 O.B 0.8 1 1.2 1.4 1.6 1.B 2

r. i I I I 1 1 1 1 1
0.2 D.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
0

0.5

-0.5

i i I I L J l_

1Q D.2 0.4 0.6 0.8 1 1-2 14 1.6 1.8 2


Time (sec)

Fig. 8.25 Control Efforts

page 183
Chapter 8 Control of the MIMO Model

A plot of the total energy (Fig. 8.26) with time shows that the total energy shoots up to
89.5 Joules due to the control actions taking place in the initial stage. The total energy is
then converted into dissipation and it decreases along a smooth curve to attain the steady

state value of 72.94 Joules.

Total Energy Dissipated


90

BB

BB

B4

82

lu 80 -

'
7B _L 4. I I I *

7B -

74

72
D 0.2 0.4 O.B O.B 1 1.2 1.4 1.6 1.B 2
Time (sec)

Fig. 8.26 Total Energy Dissipated During the Control Action

The tuning of the controller points to that a design requirement of 25 % overshoot on all

the three responses of the nonlinear model is very tight and difficult to achieve with the

dynamic equations of motion. The equations of motion maybe modified to


current

achieve very tight overshoot requirements. In terms of the settling time, the responses are

the steady states in a fairly less amount of time. The steady state values of
able to achieve

responses show that after the initial stage of landing the airplane would come to a
the
conditions on the three coordinates of the
standstill with the following equilibrium

system:

9. =
0.85
; zx = 2.88 m.m; z2 = -

1.08 m.m.

page 184
Chapter 8 Control of the MIMO Model

These values could be made zero after extensive design iterations involving modification
of the landing gear parameters and adjusting the weighting factors for the feedback gains.

Some of the issues like variation of the landing gear parameters and tuning the controller

subsequently are dealt in the following section, but further issues like modification of the

equations of motion are not considered in this study.

8.5 Performance of the Controller to Changes in Plant Dynamics

8.5.1 Variations in Initial Conditions

The overall procedure of control system design has been performed with zero initial
5
conditions for all the states except for xx(0) =
9(0) = = 0.8072 rad. A study of the

controller performance is carried out with change in initial condition of the variable 9,
whereas keeping the initial conditions for other states the same (Fig. 8.27). Let us vary

the initial conditions on 9 to 0, 2 , and 7.

Responses of th, z1 ,
and z2

Fig. 8.27 Controller Behavior in Various Initial Conditions

page 185
Chapter 8 Control of the MIMO Model

The responses show that the controller is robust enough to


bring the responses to the
same original state values. This is due to the integral
steady action on the error signals of

the states of the system. It can be seen that as we increase the initial condition on 9 to a

higher value, the overshoot increases but in lower


practical landing scenarios values of

the pitch angle are preferred as the aim the pilot is to


of
keep the nose of the plane lower
to decrease the landing speed [23].

8.5.2 Variations in Landing Gear Parameters

In order to study the effects of varying the landing gear parameters on the controller

performance, the simulations are performed with various values of the spring constants

km, and kn; the damping constants cx, and c2; and the landing speed v0. The various cases

are described below:

(a) Variations in km, and kn

The following plot (Fig. 8.28) shows the performance of the controller with different
values of km, and kn.

Responses of th. z1 . and z2


0.1 i
-

km=kn=1O0 N/m
km=kn=10O0 N/nri
km=kn=15000 N/m
0.06
i !
\ j
i i i i i
05 1 5 2.5

0.02

0
V
-0.02

-0.04

0 OS 1.5 2.5
10"

x
1 l r i l

rr o

-5
A.^
i
-
~~

s^~-~^

\y
-10

i i i i
-15

0.5 1 5 2.5

Fig. 8.28 Controller Performance under Various Values of km and kn

page 186
Chapter 8 Control of the MIMO Model

The above simulation was performed by keeping the weighting factors for the states and

integral error states at their original values designed for the case when km kn = 8000

N/m. It can be seen that as we increase the spring constants of the wheel springs, the

suspension becomes more stiffer leading to a lower pitch angle and larger steady state

values for zx, and z2. It would need a very high stiffness value for km, and kn to bring 9,
and z2 to zero but it should be noted that zx increases in the positive direction with

increase in km, and kn.

(b) Variations in cx, and c2

The following plot (Fig. 8.29) shows the performance of the controller under various

values of the damping constants cx, and c2.

Responses of th, z1 ,
and z2

0.1
c1=c2=5 Ns/m
-

d=5Ns/m; c2=20 Ns/m

\
d=c2=50 Ns/m

0.05

\ ^
i i i i

Fig. 8.29 Controller Performance under Various Values of cx and c2

page 187
Chapter 8 Control of the MIMO Model

Even though not much variation is observed in the controlled responses, the overshoot

seems to decrease with increase in the damping constants. This is quite reasonable

because by increasing the damping constants we are increasing the inherent damping in

the system. But we can not increase the damping to very high values, as it may require

powerful dampers.

(c) Variations in landing speed, vq

Simulations of the model are performed with various landing speeds and the responses

are shown in Fig. 8.30. Even though the FAA (Federal Aviation Administration) requires

[24] that a transport type aircraft be able to withstand the shock of landing at 10 ft/s at

the design landing weight, the design of the control system for the landing gear model

has been done at very high landing speeds i.e. at 10 m/s. The term vq has not shown up

in the equations of motion due to the nature of Lagrange's formulation used in deriving
the equations of motion. As far as the simulation of the landing is concerned, the landing
speed is playing a role only in the case of the input signals. Let us study the behavior of

the model by varying the landing speeds.

Responses of th, z1 ,
and z2

0 1 1I 1 1
*Q=1 m/s

vO=3.048 m/s
-

\O=10 m/s

O.-0.05

V/ i ! i

] 0 5 1 1 5 :I 2.5 3

0 02

-5 0 01

1.01 2.5
0 0.5 1.5
10"

x
i

-10

-15
1.5 2.5
0.5

Fig. 8.30 Simulation with Different Values of v0

page 188
Chapter 8 Control of the MIMO Model

A close look at the controller behavior under various


landing speeds reveals that the

model is not susceptible to changes in landing speed. This is predictable since the

controller uses feedback gains which are proportional to the landing speed.

On the basis of the control design analysis provided in this Chapter, the feedback control

design employing Linear Quadratic Regulator method along with the error space

approach has been found to provide the optimum performance requirements.

8.6 Control of the Actual Landing Scenario

The control design procedure outlined in this Chapter can be extended to the actual

landing scenario discussed in Chapter 6, where we derived a 2-dof model to represent the

A feedback loop is designed to the initial stage


landing touchdown case. simple control

which begins when the main wheels touch the runway and continues until the nose wheel

touches the runway. The two outputs are then merged using the merge blocks. Let us

assume that the landing speed, v0 = 10 m/s, and the distance traveled by the main

wheels until the nose wheel touches the runway, d =


10 m. Therefore the time required

to travel this distance is:

td = -
= lsec. (8.74)
^o

landing has to be simulated for


Hence the simulation representing the touchdown stage of

1 sec. To design a controller for the initial stage, the 2-dof model is linearized in

Simulink and a set of feedback gains for the 4 states of the system is determined by the

same procedure used for calculating the feedback gains for the 3-dof model. The

Matlab codes show the calculation of the feedback gains with the state space
following
form of the 2-dof model.

Design for 2-
the Feedback Gains for the Error State Space
% Codes to Evaluate

dof Touchdown Model

I.C on Theta=0.0873 rad


% State Space Model in Generic Form for

A=[ 0 0 1 ;
0 0 0 1;
-0.0037908
-0.01984
0
-6.0653

-674.41 -0.42151 -0.013719 0];

B=[ 0 0;
0 0;

page 189
Chapter 8 Control of the MIMO Model

6.0653 1.7987e-005
674.41 -0.998];

C=[ 1 0 0 o;
0 1 0 01;
D=[0 o;
0 01;

% To Find the State Space Model for the Error Space


system

om=diag( [1 , 1]);
-omegas*

Ae=[A, zeros(4,2);C, om]


Be=[B; D] ;

rho1=5000; % The weighting factor for zOd


(edot)
rho2=5000; % The weighting factor for zO
(edot)

P=35; % The weighting factor for theta


q=500; % The weighting factor for z1

0=
[ p 0 0 0 0 0;
0 q 0 0 0 0;
0 0 0 0 0 0;
0 0 0 0 0 0;
0 0 0 0 rhol 0;
0 0 0 0 0 rho2];

R=diag([0.01, 0.01]);

K11=lqr(Ae,Be,Q,R);

K11 =

-7.7690e-001 2.2360e+002 1.6199e-001 8.1285e-001 2.1452e-005 2.3853e-003


5.9169e+001 -8.4308e-001 1.1254e+002 -1.0133e+000 4.7842e-009 -3.5298e-006

t=0:0.01:1;
x0=[5*pi/180, zeros(1,5)];
u1=a*sin(omega*t) ;
u2=-a*omega/v2*sin(omega*t) ;
u=[u1', u2'];

% Gains for the Error Signals

K11_zO=K11(1,5:6);
K11_th_zO=K11_zO(1);
K11_z1_zO=K11_zO(2);

K1 1_z0d=K1 1(2,5:6);
K1 1_th_z0d=K1 1_z0d ( 1 );
K11_z1_z0d=K11_z0d(2);

In determining the optimal values of the gains, the variable representing the pitching

moment, Pm is manipulated such that the pitch angle follows a smooth curvilinear path

representing the smooth nose down pitching motion of the plane. A value of Pm =
0.18

page 190
Chapter 8 Control of the MIMO Model

N m is selected after trial and error to suit the values of the feedback gains calculated

for the initial stage of landing. The block diagrams representing the control of the

complete landing simulation is shown in Fig. 8.31.

th thdot z1 zldot
Control of an Actual Landing
Data StorBata StorSaia StoGeta Store
Memory3Memory2 MemoryMemoryl

h Merge

zOdotdot

H Merge zldot

zD model

zldot

Merge
Inputs

fc
Merge thdot

thdot

z2dot

z1
Data Store
Read2pata Store
thdot
Read
zldot
Data Store
Data Store Read3
Feedback Control Read I

page 191
Chapter 8 Control of the MIMO Model

Input Feedback Loops

KU
u zOdl

Input Scaling zOdl

KID
u z01

zOdotdot

Input Scaling zOd


ufb zOd
I L

zO u_zO

i zl z2 JL u_z02

_
I r}
a
Terminator
time

th
2h

z1 z2
Input Scaling zO
ufb zO

th
mode&"^

zh
z1 j
z2
Input Scaling zh
CD
ufb zh

td>=vO
Logic Box

page 192
Chapter 8 Control of the MIMO Model

2-DOF Subsystem Model (t<dM)) 0

ODE of Thdotdot

m
*/[th] I
<a

on of Thdot< Integration of zldotd >t Integration ofz2dotd


2L_JV^>>r4^) l^
\^ thdot
z2dol
z2dot
zldot

^<[*hdoq| *<[z2dot]|
tx;[zidotJ

ODE of zldotdot Derivative ODE ot22dotdot


ODE of Thdotdoi

page 193
Chapter 8 Control of the MIMO Model

State Feedback Loops

-CD
? th
CT>-
CD
ufb zO ScopeO z1

-CD
<T>- K"u z2

ufb_zOd CD
Matrix thdot
CD- Cain -CD
zldot
ufb zh
CD
z2dot

Input Feedback Loop


fort<d/vO

CD

CD
I I
..copel

1
I) ?
s CD
u zO
Integratorl

/npu* Feedback Loop


fort<dM>

I l
> cupel

X 1
W s CD
u zOd
Integratorl

page 194
Chapter 8 Control of the MIMO Model

Input Feedback Loop


t>=d/vO

CD
I I
i
ieopel
\ 1
ft.
r *
s CD
u zO
Integratorl

Input Feedback Loop


t>=d/vO

u zOd

page 195
Chapter 8 Control of the MIMO Model

Input Feedback Loop


t>=dM>

Fig. 8.31 Simulink Block Diagrams of the Actual Landing Scenario

Two separate feedback loops are required to control the system in both the cases of

landing. The responses of the overall landing simulation are plotted in Fig. 8.4 and Fig.

8.32.

page 196
Chapter 8 Control of the MIMO Model

Landing Responses in Touchdown and Taxiing


0.1

- 0.05 -

0 0.5 1.5 2.5 3.5

0 =

-10

0 0.5 1.5 2.5 3.5

I I I I I I

1 -

I
-
-1
\

ill!
0.5 1.5 2 2.5 3.5
Time (sec)

Fig. 8.32 Landing Responses in Touchdown and Taxiing

It can be seen that 9 is following the smooth curvilinear path followed by the nose of the

plane in the nose down pitching motion. Since the variable zx is a function of 0, it is also

but a smaller amplitude until the nose wheel hits the


following the same path with

runway. At this point, the variable z2 is introduced into the system producing a positive

overshoot for z2. At the same instant of time the variable zx is found to have an

undershoot. Therefore the overall situation means that when the nose wheel hits the

is immediate lift at the nose of the plane causing the variables zx to be


runway, there an

for instant of time. Once the transients are over, the


negative and z2 to be positive a small

page 197
Chapter 8 Control of the MIMO Model

three variables settle down to their respective


steady state values. The variations of the
other three states are shown in Fig. 8.33.

Landing Responses in Touchdown and Taxiing


0.1

^ 0

-0.1 j J\___jJ L 1 ! i J

~ -
-0.2

-0.3

0.5 1.5 2.5 3.5

0.4 1

0.2

r
E
o
[V
-

-0.2

-0.4

0.5 1.5 2.5 3.5

I L I I 1 1
2 -

! L I ' 1 <

o
-a
I7N

0.5 1.5 2.5 3.5

Fig. 8.33 Responses of 9, zx, and z2

page 198
Chapter 9

CONCLUSIONS

The SAE Heavy lift airplane built by the Aerodynamics design team of RIT was chosen

as the prototype model for this simulation and control system design procedure. A
lumped-parameter graphical representation of the prototype model was created with

dimensions measured from the actual model. Initially a 1-dof model considering only the
pitch degree of freedom was derived using Lagrange's equations. In this model, the
wheels of the plane were modeled as massless points moving on the runway. The

derivation of the simplified model was very useful and essential in analyzing the total

energy of the system which gave a sense of the actual dynamics represented by the

equation of motion. FFT analyses were performed on the response signals to identify the
inherent frequencies present in the response signals. A comparison of the responses with

and without the runway profile indicated that the addition of the runway profile did not

change the basic characteristics of the system as it was expected.

A more generalized model of the system was derived in Chapter 5 including two more

degrees of freedom; namely those introduced by the displacements of the centers of the

nose and main wheels. The 3-dof model was simulated in Simulink to obtain the three

responses of the system. The 3-dof model was updated to include the effects of the

runway profile and subsequently FFT analyses of the responses were performed. The FFT

analyses were very useful in studying the dominant frequencies present in the three

signals. The equations of motion derived using Lagrange's equations intensified the

effectiveness of the method in terms of its simplicity and straight forwardness in applying

to complex dynamic systems.

In order to achieve control on the system damping was introduced in Chapter 6, where the
system of the prototype model had to be modified to include a linear damper
landing gear
with its nose wheel mounted spring. It also became necessary to add a linear
along
damper to the main wheel suspension system because during an actual landing case, the

main wheels are the first elements to dissipate the dynamics during the touchdown

impact. Equations of motion of the damped system were derived using Lagrange's

equations and the responses were obtained after Simulating the system in Simulink. A

discussion on the actual landing case was produced, which involves two cases: 1) initial
later A simplified 2-dof model to represent the initial
touchdown, and 2) landing runs.

stage was derived and combined with the original 3-dof model using merge commands.

page 199
Chapter 6 Conclusions

The control design procedure outlined the various aspects needed to be considered while

designing a controller. Initially, the simplified 1-dof model was linearized in Simulink
and three types of controllers were designed using classical design techniques based on

root locus and transfer functions. The method of linearization of the nonlinear model was

carried out in order to study the dynamics in terms of the characteristic equations of the

linearized models. Three types of controllers were designed using the sisotool of Matlab

for the linearized 1-dof model. These controllers were then applied on the nonlinear 1-dof

model and found to be successful in maintaining the performance requirements.

A followed in for the 3-


more effective and powerful method was designing a controller

dof model using the state space design techniques. A 6th-order state space model was

created by linearizing the nonlinear 3-dof model in Simulink. Two methods of designing
the feedback gains were discussed such as the pole place or eigen assignment method and

the Linear Quadratic Regulator method. The second method was found to be more

effective in terms of the convergence of the solutions as this method is based on the

principle of minimization of a cost function. Moreover, it gave the user more freedom to

determine the weighting matrices Q, and R as opposed to manipulating the closed loop
eigen values as in the case of the pole placement method. Controller design was first

carried out on the linearized state space model of the original nonlinear 3-dof model.

Simulations with step inputs were carried out with integral control on the nonlinear model

and the responses were compared with those of the linearized model. The comparison

indicated a close match between the two and also proved that the linearized model was a

good approximation of the actual nonlinear model in terms of the basic


very
characteristics of the system.

inputs was carried out and modified to


A more robust approach to introducing sinusoidal

the nonlinear model. This error space approach provided integral


suit the requirement of

controls of the error signals and added 3 additional states to the already 6th-order plant

model. Out of the two types of the controllers designed for this specific example, the

with error space approach was found to be the most effective. The superior
LQR design

the controller is due to the integral action on the error signals. A study of
performance of

that the controller is efficient even with variation of the


the controller performance proved

landing gear parameters and the landing


speed.

The controller designed for the 3-dof model was applied on the actual landing simulation

design for this combined system consisted of two


discussed in Chapter 6. The control

parts: 1) initial touchdown stage, and 2) later part of landing run or taxiing. To reasonably

page 200
Chapter 6 Conclusions

describe the touchdown case, a variable for the nose down pitching moment was defined

and manipulated until a smooth response of the pitch angle was obtained in the initial
stage. It became necessary to design a simple controller to control the 2-dof model

describing the initial stage of the landing. Then the two responses were successfully
merged to obtain the overall landing responses.

To further improve the landing responses, other combinations of the weighting factors

could be tried or the equations of motion could also be modified by studying the
controlled responses of the present model. Another important factor influencing the

accuracy of the developed model is the availability of an accurate measurement or data

describing the actual prototype model. In the present case, the data measured might not

have been accurate due to manual measurement error.

The overall approach in formulating the problem, deriving the mathematical differential

the
equations representing the dynamics of the system, and achieving the control of

original nonlinear system proved to be a very successful approach in systematically

finding solutions to complicated problems.

page 201
Chapter 10

RECOMMENDATIONS FOR FUTURE WORK

The control of the landing gear system of the SAE Heavy lift Airplane showed that the

prototype model could be modified to provide a damper in the nose wheel suspension and

additional damping in the main wheel. Since the center of gravity of the plane was at the

rear end, the derived equations of motion could not provide zero equilibrium values for
all the states of the system. This maybe taken into consideration in
effectively distributing
the weight of the plane to yield zero state values in static equilibrium conditions.

Actuator and sensor dynamics need to be considered in applying the controller to the

practical scenario. The three control inputs mentioned in Chapter 8 should be studied in
order to calculate the actuator forces. The knowledge of the actuator forces is necessary in

selecting the type of actuators required in controlling the system in practical

implementation.

A study of the controlled responses point to the need that the design has to be modified to

meet tight design requirements in terms of overshoot. As the successful design comes

only after many iterations of the design procedure, there is always room for improving the
equations of motion and subsequently modifying the controller.

An experimental set up maybe developed to test the designed controller before applying it

on the actual prototype model. The testing of the controller is very essential because the

actual working conditions may differ from the theoretical simulation background and

thereby reveal some design loop holes which might have to be sorted out before the

successful implementation of the controller.

The aircraft in real life makes use of oleo-pneumatic shock absorbers which are nonlinear

in behavior. To accurately model the real time situation of landing, an oleo-pneumatic

shock absorber model maybe used in place of the linear damper considered in this

analysis.

To precisely describe the system, parameters like tire deflection, swivel angle, strut

deflection, damper-linkage strain, and airframe motion will have to be incorporated into

the model.

page 202
REFERENCES

[1] D. Yadav, R. P. Ramamoorfhy, Nonlinear Landing Gear Behavior at Touchdown,


Journal of Dynamic Systems, Measurement, and Control, Dec'91, Vol. 113.

[2] ALLSTAR Network, http://www.allstar.fiu.edu/aerojava/flight54.htm, Section


5.4-Landing.

[3] Norman S. Currey, Aircraft Landing Gear Design: Principles and Practices,
AIAA EDUCATION SERIES.

Jocelyn L. Pritchard, An Overview of Landing Gear Dynamics, NASA/TM-1999-


[4]
209143, ARL-TR-1976.

[5] Moreland, W.J., Landing Gear Vibration, AF Technical Report No. 6590, October
1951.

[6] W. Kruger, I. Besselink, D. Cowling, D. B. Doan, W. Kortum, and W. Krabacher,


Aircraft Landing Gear Dynamics: Simulation and Control, Vehicle Dynamics, 28
(1997), pp. 119-158.

[7] Josef S. Tbrok, Analytical Mechanics: with an Introduction to Dynamical

Systems, Wiley Publishers 1999.

[8] R. C. Hibbeler, Engineering Mechanics: Statics, Eighth Edition, Prentice Hall

Robert L. Woods, Kent L. Lawrence, Modeling and Simulation of Dynamic


[9] and

Systems, Prentice Hall 1997.

[10] Ned J Lindsley, and Nitin B. Talekar, A Tire Model for Air Vehicle Landing Gear
Multi-
Dynamics, Air Force Research Laboratory and Mechanical Dynamics, Inc.
45433-
Disciplinary Technologies Center and Consultant Services WPAFB, OH
7531 and Ann Arbor, MI 48105.

[11] Aviation Library, http://www.faatest.com/books/FLT/Chapter9/Touchdown.htm,


Section on Touchdown.

ALLSTAR Network, http://www.allstar.fiu.edu/aero/TOiScLand.htm, Flight


[12]
Performance -
Level 3, TakeOff and Landing.

page 203
[13] Fokker F27-600 Friendship, G-CHNL: Appendix B, Aircraft Accident Report No:
2/2000 (EW/C99/1/2), Aerodynamics.

[14] Gene F. Franklin, J. David Powell, and Abbas Emami -

Naeini, Feedback Control

of Dynamic Systems, Prentice Hall Fourth Edition.

[15] Richard C. Dorf, and Robert H. Bishop, Modern Control Systems, Addison -

Wesley Publishing Company Seventh Edition.

[16] Control System Class Handout on Root Locus Sketching Rules by Mark H.
Kempski.

[17] Simulink Tutorial Section on Chapter 1, LTI Viewer, Matlab R12.

[18] GUI Reference on Chapter 2, SISO Design Tool of Matlab R12.

[19] Ching-Fang Lin, Advanced Control System Design, American GNC Corporation,
PTR Prentice Hall, Englewood Cliffs, New Jersey 07632

[20] Uy-Loi Ly, Design of Automatic Control Systems, AA-EE-449, Department of

Aeronautics and Astronautics, Box 352400, University of Washington, Seattle,


WA 98195.

[21] Control Tutorials for Matab, Tutorials on State Space Controller Design,
http://www. engin. umich. edu/group/ctm/examples/pend/invSS. html, Carnegie

Mellon University.

[22] Handout on Using Simulink and Stateflow in Automotive Applications, The

Mathworks, Inc., 3 Apple Hill Drive, Natick, MA 01760-2098 USA.

[23] Glen B. Gilyard, and Alexander Bolonkin, Optimal Pitch Thrust-Vector Angle
and Benefits for all Flight Regimes, National Aeronautics and Space

Administration, Dryden Flight Research Center, Edwards, California 93523-0273.

Handout, http://www.aoe.vt.edu/~mason/MasonJ/M96SC05.pdft Chapter 5,


[24]
Shock Absorber Design.

page 204
Appendix A

DERIVATION OF 1-DOF MODEL

A.l Energy Derivation

The Maple codes for deriving the total energy of the 1-dof model with the runway profile

are shown below. To avoid confusion, only the relevant outputs are shown (in italics).

>
restart;

> # Analysis with road perturbations

> # KINETIC ENERGY TERMS

>KE:=(M/2)*((v-QF*cos(th)*thdot-QG*sin(th)*thdot)A2+(z0dot-

QF*sin(th)*thdot+QG*cos(th)*thdot)A2)+(IG/2)*thdotA2;

>Cphi_th:=cos(phi-th)=((QF-PD)*cos(th)+QP*sin(th))/ED;

> phi:=solve(Cphi_th,phi);

> Cphi:=expand(cos(phi));

Cphi:=cos(th)A2*QF/ED-cos(th)A2*PD/ED+cos(th)*QP*sin(th)/ED-sin(th)*sqrt(1-

cos(th)A2*QFA2/(EDA2)+2*cos(th)A2*QF*PD/(EDA2)-2*cos(th)*QF*QP*sin(th)/(

cos(th)A2*PDA2/(EDA2)+2*cos(th)*PD*QP*sin(th)/(EDA2)-QPA2*sin(th)A2/(EDA2))

>Uspr:=expand((ks/2)*(BD-DC*Cphi-Ls)A2);

>Ugrav:=M*(zO+QF*cos(th)+QG*sin(th))*g;

> PE1:=Ugrav+L)spr;

PE1
ks*DCA2*sin(th)A3*cos(th)*QF*QP/(EDA2)-

ks*DC*cos(th)A2*QF*BD/ED-

ks*DCA2*cos(th)A4*QF*PD/(EDA2)+1/2*ks*DCA2*cos(th)A4*QFA2/(EDA2^^

PDA2/(EDA2)+ks*DC*sin(thysqrt(1-cos(th)A2*QFA2/(EDA2)+2*cos(th)A2*QF*PD/(EDA2y

page 205
Appendix A Derivation of 1-dofFreedom Model

QPA2*sin(th)A2/(EDA2))*BD-1/2*ks*DCA2*sin(th)A4*QPA2/(EDA2)-ks*DC*sin(th)^
cos(th)A2*QFA2/(EDA2)+2*cos(th)A2*QF*PD/(EDA2)-2*cos(th)*QF*QP*sin(th)/(EDA2)-

cos(th)A2*PDA2/(EDA2)+2*cos(th)*PD*QP*sin(th)/(EDA2)-

QPA2*sin(th)A2/(EDA2))*Ls+ks*DCA2*cos(th)A3*QF*QP*sin(th)/(ED^
D-ks*DCA2*cos(th)A3*PD*QP*sin(th)/(EDA2)+ks*DCA2*cos(th)A2*PD*sin(th)*sq
cos(th)A2*QFA2/(EDA2)+2*cos(th)A2*QF*PD/(EDA2)-2*cos(th)*QF*QP*sin(th)/(EDA2)-

cos(th)A2*PDA2/(EDA2)+2*cos(th)*PD*QP*sin(th)/(EDA2)-QPA2*sin(th)A2/(EDA2))/ED-

ks*DC*cos(th)A2*PD*Ls/ED-

ks*DC*cos(th)*QP*sin(th)*BD/ED+1/2*ks*DCA2*cos(th)A2*QPA2*sin(th)A2/(EDA2^^

ks*DCA2*cos(th)*QP*sin(th)A2*sqd(1-cos(th)A2*QFA2/(EDA2)+2*cos(th)A2*QF*PD/(E^

2*cos(th)*QF*QP*sin(th)/(EDA2)-cos(th)A2*PDA2/(EDA2)+2*cos(th)*PD*^^

QPA2*sin(th)A2/(EDA2))/ED+ks*DC*cos(th)*QP*sin(th)*Ls/ED-

ks*DCA2*cos(th)A2*QF*sin(th)*sqd(1-cos(th)A2*QFA2/(EDA2)+2*cos(th)A2*QF*PD/(EDA2y
2*cos(th)*QF*QP*sin(th)/(EDA2)-cos(th)A2*PDA2/(EDA2)+2*cos(th)*PD*Q^

QPA2*sin(th)A2/(EDA2))/ED+ks*DC*cos(th)A2*QF*Ls/ED-

1/2*ks*DCA2*sin(th)A2*cos(th)A2*QFA2/(EDA2)-

ks*BD*Ls+M*(zO+cos(th)*QF+QG*sin(th))*g+1/2*ks*DCA2*sin(th)A2

>
PE2:=taylor( PE1, th=0, 3 );

> PEquad:=convert(PE2, polynom);

> Energy:=KE+PEquad;

> # Data in SI system

> QF:=0.165 : QG:=0.0127 : QP:= 0.3048 : DC:=0.035 : PD:=0.132 : ED:= 0.046 : BD:=0.063 :

> M:= 11.84 : g:=9.81: IG:=1.966:

> A:=0.0025 : lambda:=1 : Ls:=0.03 : v:=


10 :

> ks :=
5676.7088;

> zO:=A*sin(2*Pi*v*t/lambda);

> z0dot:=diff(z0,t);

> Energyl :=simplify(KE+PEquad);

page 206
Appendix A Derivation of 1-dofFreedom Model

Energyl := 61 1.341 5680-1 9. 53600000*cos(th)*thdot-

1.503680000*sin(th)*thdot+1.145126837*thdotA2+.1460701452*cos(62.83185308*t)A2-

.3068707704*cos(62.83185308*t)*sin(th)*thdot+.23

1 *cos(62. 83 1 85308*t) *cos(th) *thdot+. 2903 760000*sin(62. 83 1 85308*t)-

7.821 51 7040*th+1 03.0881 973*thA2

> simplify(eval(Energy1
,{th=Pi/8,thdot=0,t=0}));

624.3136312

> with(plots):

>
implicitplot3d( {Energyl =624.31 3631 3}, th=-0.7..0.81, thdot=-0.65..18, t=0..Pi, axes=boxed,

color=grey, labels=[th,thdot,t]);

> animate3d(Energy1,th=-Pi/8..Pi/8, thdot=-5..20, t=0..Pi, frames=50, axes=boxed);

>
Energy_no_t:=subs(t=0, Energyl);

>
contourplot3d( {Energy_no_t}, th=-Pi/8..Pi/8, thdot=-0.5..15, axes=boxed, contours=15,
color=
black);

> fieldplot3d([th,thdot,t], th=-Pi/8..Pi/8, thdot=-0.5..15, t=0..Pi, axes=boxed, arrows=SLIM,

color=
black);

> gradplot3d(Energy1, th=-Pi/8..Pi/8, thdot=-5..20, t=0..10, grid=[5,5,5], color=blue, axes=boxed);

>
plot3d( {Energy_no_t}, th=-Pi/8..Pi/8, thdot=-5..20,

color=[sin(th*thdot),cos(th*thdot),tan(th*thdot)], axes=boxed, contours=1 5);

> PEquad;

1 9.341 56797+.29037600*sin(20*Pi*t)-7.821 51 704*th+1 03.0881 973*thA2

> plot3d(PEquad, th=-Pi/8..Pi/8, t=0..Pi, axes=boxed, contours=20, color=grey);

A.2 Derivation of Equation of Motion for 1-DOF Model

1-dof model the runway


The Maple codes for deriving the equation of motion for with

profile are shown below. The outputs are shown in italics.

page 207
Appendix A Derivation of 1-dofFreedom Model

>
restart;

>Cphi_th:=cos(phi-th)=((QF-PD)*cos(th)+QP*sin(th))/ED;

>
phi:=solve(Cphi_th,phi);

>
Cphi:=expand(cos(phi));

>Uspr:=expand((ks/2)*(BD-DC*Cphi-Ls)A2);

>Ugrav:=M*(zO+QF*cos(th)+QG*sin(th))*g;

>
PE1:=Ugrav+Uspr;

>PE2:=taylor(PE1,th=0, 3);

> PEquad:=convert(PE2, polynom);

> # KINETIC ENERGY TERMS

>KE:=(M/2)*((v-QF*cos(th)*thdot-QG*sin(th)*thdot)A2+(z0dot-

QF*sin(th)*thdot+QG*cos(th)*thdot)A2)+(IG/2)*thdotA2;

> L:=KE-PEquad;

> # Differential w.r.t theta

> Lth:=diff(L,th);

> Lthdot:=diff(L,thdot);

> # Lthdot redefined as a function of t

>Lthdot_t:=subs(th=th(t),zO=zO(t),Lthdot);

> thdot:=diff(th(t),t);

> zOdot:=diff(zO(t),t);

> LthdotJ;

> # Time Differential of thdot

page 208
Appendix A Derivation of 1-dofFreedom Model

>
tLthdot:=diff(Lthdot_t,t);

> # EOM in theta co-ordinate

> Eqn_mot1 :=subs({th(t)=th,zO(t)=zO},{tLthdot-Lth=0});

> # Data in SI system

> QF:=0.165 : QG:=0.0127 : QP:= 0.3048 : DC:=0.035 : PD:=0.132 : ED:= 0.046 : BD:=0.063 :

> M:= 11.84 : g:=9.81: IG:=1.966:

> A:=0.0025 : l:=1 : Ls:=0.03 : v:=


10 :

> ks :=
5676.7088;

> Eqn_mot1 :=subs({th(t)=th,z1 (t)=z1 ,z2(t)=z2,zO(t)=zO,zh(t)=zh},{tLthdot-Lth=0});

Eqn_mot1 :=
{1 1.84000000*(.165*sin(th)*diff(th,t)A2-. 165*cos(th)*diff(th, '$'(t,2))-. 127e-

1*cos(th)*diff(th,t)A2-. 127e-1*sin(th)*diff(th, T(t,2)))*(-. 1 65*cos(th)-. 127e-1*sin(th))-

7.8215170+11. 84000000*(diff(z0Z$'(t,2))-.165*cos(th)*diff(th,t)A2-.165*sin(th)W^

. 127e-1*sin(th)*diff(th,t)A2+. 127e-1*cos(th)*diff(th, -$'(t,2)))*(-.165*sin(th)+. 127e-

1*cos(th))+1.966*diff(thZ$'(t,2))+206.1763946*th =
0}

> # Simplification of the EOM for 2 DOF Model

> # Equation of Motion Rewritten as a Function of Time

>Eqn_mot1_0:={11.84000000*(.165*sin(th(t)rdiff(th(t),t)A2-.165*cos(th(t))*diff(th(t);$'(t,2))-

1*sin(th(t)))-7.8215169+11.84000000*(diff(z0(t),-$'(t,2))-.165*cos(th(t))*diff(th(t),t)A2-

. 1 65*sin(th(t))*diff(th(t),T (t,2))-. 1 27e-1 *sin(th(t))*diff(th(t),t)A2+. 1 27e-

1 *cos(th(f))*diff(th(f),T (t,2)))*(-. 1 65*sin(th(t))+. 1 27e-

1*cos(th(t)))+1.966*diff(th(t),-$"(t,2))+206.1763944*th(t) =
0};

> EOM1_0:=simplify(Eqn_mot1_0);

EOM1_0 :=
{2.290253674*diff(th(t), '$'(t,2))-7.821516900-
1.953600000*diff(z0(t), '$'(t,2))*sin(th(t))+. 1503680000*diff(z0(t), -$-(t,2))*cos(th(t))+206.1763944*t

h(t) =
0}

page 209
Appendix A Derivation of 1-dofFreedom Model

>solve(EOM1_0,diff(th(f),'$'(t,2)));

{diff(th(t), '$'(t,2)) =
3.415131253+.8530059452*diff(z0(t), '$'(t,2))*sin(th(t))-.6565560912e-

1*diff(z0(t)Zr(t,2))*cos(th(t))-90.02338769*th(t)}

>
zO:=A*sin(2*Pi*v*t/l);

>
zOdot:=diff(zO,t);

>
zOdotdot:=diff(zOdot,t);

A.3 Codes for Running FFT on the Response Signal

The codes written in Matlab to run FFT on the response signal with the runway profile are

shown below:

% WAVEFORM GENERATION FOR 1622 POINTS ON THETA

[N,p]=size(time 1 );

T=l/12.8;

k=0:N-l;

figure(l);subplot(2,l,l);plot(timel,thl);grid

title('Generated waveform for Nl points');

xlabel('Time(sec)');ylabel('Theta');

subplot(2,l,2);plot(thl,thdotl);grid

title('Plot ofThdot vs. Theta');

xlabel('Theta');ylabel('Thdof);

% RUNNING FFT ON th

magthl=abs(ffl(thl));

[N2,q]=size(magthl);

page 210
Appendix A Derivation of 1-dofFreedom Model

hertz=k*(l/(N*T));

figure(2);subplot(2, 1 1 );stem(hertz( 1 :N/2),magthl ( 1 :N/2));grid;


,

title('Magnitude of Theta plotted against Hz');

xlabelCHz'XylabelCTheta');

subplot(2, 1 ,2);plot(hertz( 1 :N/2),magth 1 ( 1 :N/2));grid;

title('Magnitude of Theta plotted against Hz');

xlabelCHz'^ylabelCTheta');

[y,y 1 ]=max(magth 1 )

figure(3 );subplot(2, 1 1 );stem(k( 1 :N/2),magth 1 ( 1 :N/2));grid;


,

title('Magnitude of Theta plotted against k');

xlabelCk');ylabel('Theta');

subplot(2,l,2);plot(k(l:N/2),magthl(l:N/2));grid;

title('Magnitude of Theta plotted against Hz');

xlabelCk');ylabel('Theta');

page 211
Appendix B

DERIVATION OF 3-DOF MODEL

B.l Derivation of Equations of Motion

The Maple codes for deriving the equations of motion of the 3-dof model are shown

below. To avoid confusion only the important outputs are shown (in italics).

>
restart;

> Cphi:=((QF-PD)*cos(th)+QP*sin(th)+z1 -z2)/ED;

> x_ks:=BD-DC*Cphi-Ls;

>Uspr:=expand((ks/2)*x_ksA2)+(km/2)*(z1-z0)A2+(kn/2)*(z2-zh)A2;

>Ugrav:=(M*(zO+z1+QF*cos(th)+QG*sin(th))+Mm*z1+Mn'z2)*g;

> PE1:=Ugrav+Uspr;

PE1 := .29037600*sin(20*Pi*t)-26. 1 783435*z1 +14.461 1691 3*cos(th)-


789430873*

41. 96948278*sin(th)+142.8388635*z2+3.090967942+33.05566849*cos(th)*sin(th)+1.

cos(th)A2+1 52. 6570872*sin(th)A2+108. 4503560*cos(th) *z1-


1*z1A2-
108.4503560*cos(th)*z2+1 001. 686923*sin(th)*z1 -1001. 686923*sin(th)*z2+1 643. 18721

3286.374422*z1*z2+1643.187211*z2A2+4000*(z1-.25e-2*sin(20*Pi*t))A2+4000*(z2-.25e-

3758643)*th-(-
2*sin(62.83185308*t+1. 71 3758643)-.4411826092e-2*cos(62.831 853081+1. 71

6865e-
.3892841
895e-2*sin(62.83185308*t+1. 71 3758643)+. 442507

1*cos(62.83185308*t+1.713758643))*thA2)A2

>PE2:=taylor(PE1,th=0, 3);

> PEquad:=convert(PE2, polynom);

> # KINETIC ENERGY TERMS

QF*sin(th)*thdot+QG*cos(th)*thdot)A2)+(IG/2)*thdotA2+(Mm/2)*z1dotA2+(Mn/2)*z2dotA2;

> # DISSIPATIVE COMPONENT

page 212
Appendix B Derivation of 3-dofFreedom Model

>xdot:=-DC*(-(QF-PD)*sin(th)*thdot+QP*cos(th)*thdot+z1dot-z2dot)/ED;

>
Fs:=-c*xdot;

>
d:=(c/2)*xdotA2;

>
cthdot:=diff(d,thdof);

>cz1dot:=diff(d,z1dot);

>
cz2dot:=diff(d,z2dot);

>
L:=KE-PEquad;

> # Differential w.r.t theta

>
Lth:=diff(L,th);

># Differential w.r.t z1

>Lz1:=diff(L,z1);

># Differential w.r.t z2

>
Lz2:=diff(L,z2);

> Lthdot:=diff(L,thdot);

> Lz1dot:=diff(L,z1dot);

> Lz2dot:=diff(L,z2dot);

> thdot:=diff(th(t),t);

>z1dot:=diff(z1(t),t);

> z2dot:=diff(z2(t),t);

> xO:=vO*t;

> zOdot:=diff(zO(f),t);

> zhdot:=diff(zh(t),t);

page 213
Appendix B Derivation of 3-dofFreedom Model

> # Lthdot redefined as a function of t

>Lthdot_t:=1/2*M*(2*(vO-QF*cos(th(t))*thdot-QG*sin(th(t))*thdot)*(-cos(th(t))*QF-

QG*sin(th(t)))+2*(z0dot+z1dot-QF*sin(th(t))*thdot+QG*cos(th(t))*thdot)*(-

QF*sin(th(t))+QG*cos(th(t))))+IG*thdot;

> # Lzldot redefined as a function of t

> Lz1 dot_t:=Mm*z1 dot+1/2*M*(2*z0dot+2*z1 dot-2*QF*sin(th(t))*thdot+2*QG*cos(th(t))*thdot);

> # Lz2dot redefined as a function of t

> Lz2dot_t:=Mn*z2dot;

> # Time Differential of thdot

>
tLthdot:=diff(Lthdot_t,t);

> # Time differential of z1 dot

>tLz1dot:=diff(l_z1dot_t,t);

> # Time differential of z2dot

> tLz2dot:=diff(Lz2dot_t,t);

> # Data in SI system

QG:=0.0127 : QP:= 0.3048 : DC:=0.035 : PD:=0.132 : ED:= 0.046 : BD:=0.063


> QF:=0.165 :

> Ls:=0.03 : rm:=0.064262 : rn:=0.033655 :

> M:= 11.84 : g:=9.81: IG:=1.966: Mm:=0.021 : Mn:=0.031 :

> ks :=
5676.7088;

> Eqn_mot1 :=subs({th(f)=th,z1 (t)=z1

127e-
Eqn_mot1 := {1 1.84000000*(.165*sin(th)*diff(th,t)A2-. 165*cos(th)*diff(th, T(t,2))-.
127e-1*sin(th))-
1*cos(th)*diff(th,t)A2-. 127e-1*sin(th)*diff(th, '$'(t,2)))*(-. 1 65*cos(th)-.
8.9138143+11.84000000*(diff(z0Zr(t,2))+diff(z1,'$'(t,2))-.165*cos(th)*diff(th,t)A2-

-$-(t,2)))*(-
.165*sin(th)*d
'$'(t,2))-. 127e-1*sin(th)*diff(th,t)A2+. 127e-1*cos(th)*diff(th,

page 214
Appendix B Derivation of 3-dof Freedom Model

.
165*sin(th)+. 127e-1*cos(th))+1.966*diff(th, '$'(t,2))+2*(143.6370717+54.2251780*z2-

54.2251 780*z1)*th+1 001. 686924*z1 -1001. 686924*z2+. 5789224953*c*(-.33e-

1*sin(th)*diff(thJ)+.3048*cos(th)*diff(th,t)+diff(z1,t)-d^

> Eqn_mot2:=subs({th(f)=th,z1 (t)=z1 ,z2(t)=z2,z0(t)=z0,zh(t)=zh},{tLz1 dot-Lz1 +cz1 dot=0});

Eqn_mot2 :=
{1 1. 861 00000*diff(z1 '$'(t,2))+11. 84000000*diff(z0,
,
'$'(t,2))-

1. 953600000*cos(th) *diff(th, t)A2- 1 .


953600000*sin(th) *diff(th, T(t, 2))-
.1503680000*sin(th)*diff(th,t)A2+.1503680000*cos(th)*diff(thZr(t,2))+3286.374423*z1 +82.272012

5-3286.374423*z2+km*(z1-z0)-54.2251780*thA2+1001.686924*th+.5789224953*c*(-.33e-

1 *sin(th) *diff(th, t) +. 3048*cos(th) *diff(th, t) +diff(z1, t)-diff(z2, t)) =


0}

> Eqn_mot3:=subs({th(t)=th,z1 (t)=z1 ,z2(t)=z2,zO(t)=zO,zh(t)=zh},{tLz2dot-Lz2+cz2dot=0});

Eqn_mot3 :=
{.31e-1*diff(z2, '$'(t,2))+3286.374423*z2+34.3885075-3286.374423*z1+kn*(z2-
zh)+54.2251780*thA2-1001.686924*th-.5789224953*c*(-.33e-

1 *sin(th) *diff(th, t) +. 3048*cos(th) *diff(th, t)+diff(z1, t)-diff(z2, t)) =


0}

> # Simplification of the EOM for 2 DOF Model

> # Equation of Motion-1 Rewritten as a Function of Time

>Eqn_mot1_0:={11.84000000*(.165*sin(th(t))*diff(th(t),t)A2-.165*cos(th(t))*diff(th(t),-$-(t,2))-

127e-1*cos(th(t))*diff(th(t),t)A2-.127e-1*sin(th(t))*diff(th(t),-$-(t,2)))*(-.165*cos(th(t))-.127e-

1*sin(th(t)))-8.91381431+11.84000000*(diff(z0(t),-$'(t,2))+diff(z1(t),-$-(t,2))-

1*cos(th(t))*diff(th(t),-$'(t,2))r(-.165*sin(th(t))+.127e-

1*cos(th(t)))+1.966*diff(th(t),'$'(t,2))+1001.686924*z1(t)-

1001.686924*z2(t)+2*(54.2251780*z2(t)+143.6370718-54.2251780*z1(t))*th(t)+.5789224953*c*(-

1*sin(th(t))+.3048*cos(th(t))) =
0};

> EOM1_0:=simplify(Eqn_mot1_0);

>P1:=solve(EOM1_0,diff(th(t),-$'(t,2)));

{diff(th(t), '$'(t,2))
=
-437.369421 3*z1(t)+437.3694213*z2(t)+3.8920641 90-1 25.4333294*th(t)-

47.35298855*th(t)*z2(t)+47.35298855*th(t)*z1(t)-.6565560912e-

50850559406-
1 *diff(z1(t), Tfr, 2))*cos(th(t)) +. 8530059452*diff(z0(t), Tft 2)) *sin(th(t)) +.

page 215
Appendix B Derivation of 3-dofFreedom Model

2*c*sin(th(t))*diff(th(t),t)*cos(th(t))-.2752736977e-3*c*dm
1*c*cos(th(t))A2*diff(th(t),t)-.6565560912e-

1*dm(z0(t)Z^(t,2)rcos(th(t))+.8530059452*diff(z1(t)Zr(t,2)rsinm
1*c*diff(z2(t),t)*cos(th(t))-.8341627199e-2*c*diff(z2(t)^)*sin(th(t))-.77046302
1*c*diff(z1(t),t)*cos(th(t))+.8341627199e-2*c*diff(z1(t),t)*sW^^

> # Equation of Motion-2 Rewritten as a Function of Time

>Eqn_mot2_0:={11.86100000*diff(z1(t),'$-(t,2))+11.84000000*diff(z0(t),'$"(t,2))-

1.953600000*cos(th(t))*diff(th(t),t)A2-1.953600000*sin(th(t))*diff(th(t);$-(t,2))-

.1503680000*sin(th(t))*diff(th(t),t)A2+.1503680000*cos(th(t))*diff(th(t),'$-(t,2))+82.2720125-

3286.374423*z2(t)+3286.374423*z1(t)+1001.686924*th(t)-54.2251780*th(t)A2+km*(z1(t)-

z0(t))+.5789224953*c*(-.33e-1*sin(th(t))*diff(th(t),t)+.3048*cos(th(t))*diff(th(t),t)+diff(z1(t),t)-

diff(z2(t),t)) =
0};

>
EOM2_0:=simplify(Eqn_mot2_0);

>p2:=solve(EOM2_0,diff(z1(t),T(t,2)));

{diff(z1(t)Z$'(t,2))=-

.9982294916*diff(z0(t)Z$'(t,2))+.1647078661*cos(th(t))*dimh(t),t)A2+.1647078661*sin(m

h(t), '$'(t,2))+. 1 267751 454e-1*sin(th(t))*diff(th(t),t)A2-. 1267751 454e-1*cos(th(t))*diff(th(t), '$'(t,2))-


6.936347062+277.0739755*z2(t)-277.0739755*z1(t)-84.45214771*th(t)+4.571720597*th(t)A2-

.
8430992328e- 1
*km*z1(t) +. 8430992328e- 1 *km*z0(t) +. 1610694068e-2*c*sin(th(t)) *diff(th(t), t)-
. 1 48769561 3e-1*c*cos(th(t)) *diff(th(t), t)-.488089 1116e-1 *c*diff(z1(t), t) +. 488089 1116e-

1*c*diff(z2(t),t)}

> # Equation of Motion-3 Rewritten as a Function of Time

>Eqn_mot3_0:={.31e-1*diff(z2(t),'$"(t,2))+34.3885075-3286.374423*z1(t)+3286.374423*z2(t)-

1001.686924*th(t)+54.2251780*th(t)A2+kn*(z2(t)-zh(t))-.5789224953*c*(-.33e-

1*sin(th(t))*diff(th(t),t)+.3048*cos(th(t))*diff(th(t),t)+diff(z1(t),t)-diff(z2(t),t))
=
0};

> EOM3_0:=simplify(Eqn_mot3_0);

>p3:=solve(EOM3_0,diff(z2(t),T(t,2)));

{diff(z2(t), T(t,2))
=
-11 09.306694+1 0601 2.0782*z1(t)-1 0601 2.0782*z2(t)+3231 2.481 42*th(t)-
1749.199290*th(t)A2-32.25806452*kn*z2(t)+32.25806452*kn*zh(t)-

page216
Appendix B Derivation of 3-dofFreedom Model

.6162723335*c*sin(th(t))*diWh(t),t)+5.692115374*c*cos(th(t)^^

,t)-1 8.67491 920*c*diff(z2(t), t)}

>
xO:=vO*t;

>
zO:=A*sin(2*Pi*xO/l);

>
zOdot:=diff(zO,t);

zOdot :=
2*A*cos(2*Pi*v0*t/lambda)*Pi*v0/lambda

>Sphi_th:=sqrt(1-(((QF-PD)*cos(th)+QP*sin(th))/ED)A2);

>vO:=10:l:=1 : A:=0.0025 :

>xE:=xO-QF*sin(th)+QP*cos(th)+PD*sin(th)-ED*Sphi_th;

xE .
-
1 0*t-. 33e- 1 *sin(th) +. 3048*cos(th)-.46e- 1 *sqrt(1-472. 5897921 *(. 33e-

1 *cos(th) +. 3048*sin(th))A2)

>zh1:=A*sin(2*Pi*xE/l);

zh1 :=.25e-2*sin(2*Pi*(10*t-.33e-1*sin(th)+.3048*cos(th)-.46e-1*sqrt(1 -472.5897921 *(.33e-

1 *cos(th) +. 3048*sin(th))A2)))

>zh_t:=taylor(zh1,th=0, 3);

> zh_o:=convert(zh_t, polynom);

> #Redefine zh_o in terms of t

>
zh:=subs(th=th(t),zh_o);

>
zhdot:=diff(zh,t);

zhdot=. 1570796327*cos(62. 83185308*t+1. 713758643)-

2*cos(62.83185308*t+1.713758643)*diff(th(t),t)+(-

.2445944700*cos(62.83185308*t+1. 71 3758643)-
2.780357794*sin(62.83185308*t+1.713758643))*th(t)A2+2*(-.3892841895e-

page 217
Appendix B Derivation of 3-dofFreedom Model

2*sin(62. 831 85308*t+1 7 13758643)+. 4425076865e-


.

rcos(62.83185308*t+1.713758643))*th(t)*diff(th(t),t)

B.2 Matlab Codes to Verify the Simulation Done in Simulink

The following codes in Matlab compute the responses of the 3-dof model using ode-45

function.

"nonlin_model_3dof_no_z0"
% To Compare the Responses of the Simulink Model with the Nonlinear

Simulation of the System done by ode45

% Both Models Assume Zero Road Perturbations

function odesolvel

"nonlin_model_3dof_no_z0"

% To Simulate the Simulink Model using Matlab codes

[time,x] =
sim('nonlin_model_3dof_no_z0'); % Simulation Command

'x' 'th'

th_l=x(:,l); % Assigning 1st Column of to the Variable

'x' 'zl'

% Assigning 2nd Column of to the Variable


zl_l=x(:,2);

'x' 'z2'
% Assigning 3rd Column of to the Variable
z2_l=x(:,3);

'x' 'thdot'

% Assigning 4th Column of to the Variable


th_ldot=x(:,4);

'x' 'zldot'

% Assigning 5th Column of to the Variable


zl_ldot=x(:,5);

'x' 'z2dot'

% Assigning 6th Column of to the Variable


z2_ldot=x(:,6);

%To solve the Nonlinear ODE's of 3-dof System using ode45

tspan =
[0 50]; % Time Span of the Simulation of ode45

% Initial Conditions of the State Vector


sO
=
[0; 0; 0; 0; 0; 0];

[t,s]=ode45(@f,tspan,s0); % Simulation Command for ODE-45

's' 'th'

% 1st Column of to the Variable


th_2=s(:,l); Assigning

'thdot'
's'

th 2dot=s(:,2); % Assigning 2nd Column of to the Variable

page 218
Appendix B Derivation of 3-dofFreedom Model

's' 'zl'
zl_2=s(:,3); % Assigning 3rd Column of to the Variable

's' dot'
zl_2dot=s(:,4); % Assigning 4th Column of to the Variable 'zl

's' 'z2'
z2_2=s(:,5); % Assigning 5th Column of to the Variable

's' 'z2dot'
z2_2dot=s(:,6); % Assigning 6th Column of to the Variable

figure(l);

plot(time,th_l ,t,th_2,'y');grid; % Comparison of the Responses

xlabel('Time');

ylabel('Response');

title('Response of Theta Vs Time');

legend('Simulink Model Response','Matlab Codes Response');

figure(2);

plot(time,th_l dot,t,th_2dot,'y');grid; % Comparison of the Responses

xlabel('Time');

ylabel('Response');

title('Response of Thetadot Vs Time');

legend('Simulink Model Response','Matlab Codes Response');

figure(3);

plot(time,zl_l,t,zl_2,'y');grid; % Comparison of the Responses

xlabel('Time');

ylabel('Response');

title('Response of zl Vs Time');

legend('Simulink Model Response','Matlab Codes Response');

page 219
Appendix B Derivation of 3-dofFreedom Model

figure(4);

plot(time,zl_ldot,t,zl_2dot,'y');grid; % Comparison of the Responses

xlabel('Time');

ylabel('Response');

title('Response of zldot Vs Time');

legend('SimuIink Model Response','Matlab Codes Response');

figure(5);

plot(time,z2_l ,t,z2_2,'y');grid; % Comparison of the Responses

xlabel('Time');

ylabel('Response');

title("Response of z2 Vs Time');

legend('Simulink Model Response','Matlab Codes Response');

figure(6);

plot(time,z2_ldot,t,z2_2dot,'y');grid; % Comparison of the Responses

xlabel('Time');

ylabel('Response');

title('Response of z2dot Vs Time');

legend('Simulink Model Response','Matlab Codes Response');

% Function to define the Equations of Motion of the System

function dsdt =f(t,s)

c=20;

page 220
Appendix B Derivation of3-dof Freedom Model

km=8000;

kn=8000;

dsdt =
[ s(2)

(c*(s(6)-s(4))*(0.073 8*cos(s( 1 ))+0.033 *sin(s(l )))+s(2)A2*(0.


14*sin(s( 1 ))*cos(s( 1 ))-
0.022*(cos(s( 1 )))A2+0.0 1 08)+c*s(2)*(0.00 1 1 -0.0077*sin(s( 1 ))*cos(s( 1 ))-
0.024*(cos(s(l)))A2)+s(l)A2*(3.9*sin(s(l))-0.3*cos(s(l)))+s(3)*(18.19*cos(s(l))+47.35*s(l)-

236.34*sin(s(l))-437.37)+s(3)*km*0.0055*cos(s(l))+s(5)*(-18.19*cos(s(l))-

47.35*s(l)+236.34*sin(s(l))+437.37)+s(l)*(5.54*cos(s(l))-72.04*sin(s(l))-125.43)+0.455*cos(s(l))-

5.92*sm(s(l))+3.892-s(3)*lan*0.072*sm(s(l)))/(0.86+0.14*(cos(s(l)))A2+0.022*sin(s(l))*cos(s(l)))

s(4)

(s(2)A2*(0. 1 65*cos(s( l))+0.0 1 3 *sin(s(l )))+c*s(2)*(0.00 1 56*sin(s( 1 ))-

0.0039*sin(s( 1 ))*(cos(s( 1 )))A2-D.0 14*cos(s( l))-0.00054*(cos(s( 1 )))A3)+c*(s(6>


s(4))*(0.013*sin(s(l))*cos(s(l))+0.000397*(cos(s(l)))A2+0.0474)+s(l)A2*(0.2e-

009*(cos(s( 1 )))A2+0. 1 04e-0 1 0*sin(s(l ))*cos(s( 1 ))+4.572)+s(5)*(0.347e-009*sin(s( 1 ))*cos(s( 1 ))-


7.2*s(l)*sin(s(l))+72.04*sin(s(l))-5.54*cos(s(l))+277.07)+s(3)*(-0.347e-

009*sm(s(l))*cos(s(l))+7.2*s(l)*sin(s(l))-72.04*sin(s(l))+5.54*cos(s(l))-277.07-km*(0.484e-

013*sin(s(l))*cos(s(l))+0.0843))+s(l)*(-84.452+1.59*cos(s(l))-20.66*sin(s(l))-0.1998e-

009*sin(s(l ))*cos(s( 1 )))+0.64 1 *sin(s( 1 ))-0.0493 *cos(s(l ))-0.2e-009*(cos(s( l)))A2-0. 1 1 52e-

009*sin(s(l))*cos(s(l))-6.936)/(0.86+0.14*(cos(s(l)))A2+0.022*sin(s(l))*cos(s(l)))

s(6)

c*s(2)*(5.692*cos(s(l))-0.616*sin(s(l)))+18.675*c*(s(4)-s(6))-1109.31+106012.0782*(s(3)-

s(5))+32312.481*s(l)-1749.2*s(l)A2-32.258*kn*s(5)];

A plot of the comparison of the responses is shown in Fig. B. 1 .

page 221
Appendix B Derivation of 3-dof Freedom Model

Simulink Model Response


Matlab Codes Response

0.01

0.005
*'
A ^
j-\-H-z\yj
i i \ i

/ v A
4
V-J---t

-iH-
-^
-0.005
IX

J L J L
-0.01

2 3 4 5 6 7 10
Time

Fig. B.l Comparison of Responses

B.3 Matlab Codes to run FFT on the 3-dof Responses

% RUNNING FFT ON th

%Measures the no: of data points in the signal


[N,p]=size(th2)

T=1/617; %Sampling frequency 512 Hz, higher than the highest sinusoidal frequency

k=0:N-1 ; %Counter for FFT

f=fft(th2);

magth=abs(f);

freq=k*(1/(N"T));

figure(1);

page 222
Appendix B Derivation of 3-dofFreedom Model

subplot(3,1,1);

stem(freq(1:N/2),magth(1:N/2));

grid;

title('Frequency Contents in Theta, z1 & z2');

xlabel('Frequency (Hz)');

ylabel('mag[Theta]');

[y,y1]=max(magth)

t=(0:N-1)/N*2*pi;

x=cos(f);

y=sin(t);

figure(2);

subplot(3,1,1);

plot3(x,y1magth','d','fill');

grid;

xlabel('Real');

ylabel('lmaginary');

zlabel('Amplitude');

view([-65 30]);

% RUNNING FFT ON z1

f1=fft(z12);

magz1=abs(f1);

figure(1);

page 223
Appendix B Derivation of 3-dofFreedom Model

subplot(3,1,2);

stem(freq(1 :N/2),magz1 (1 :N/2));

grid;

xlabel('

Frequency (Hz)');

ylabel('mag[z1]');

[y,y1]=max(magz1)

t=(0:N-1)/N*2*pi;

x=cos(t);

y=sin(f);

figure(2)

subplot(3,1,2);

plot3(x,y,magz1','d','fill');

grid;

xlabel('Real');

ylabel('lmaginary');

zlabel('Amplitude');

view([-65 30]);

% RUNNING FFT ON z2

f2=fft(z22);

magz2=abs(f2);

figure(1)

subplot(3,1,3);

page 224
Appendix B Derivation of 3-dof Freedom Model

stem(freq(1 :N/2),magz2(1 :N/2));

grid;

xlabel('Frequency (Hz)');

ylabel('mag[z2]');

[y,y1]=max(magz2)

t=(0:N-1)/N*2*pi;

x=cos(t);

y=sin(t);

figure(2)

subplot(3,1,3);

plot3(x,y,magz2','d','fiir);

grid

xlabel('Real');

ylabel('lmaginary');

zlabel('Amplitude');

view([-65 30]);

page 225