Anda di halaman 1dari 49

L1

Augmented Controller for a Lateral/Directional


Manoeuvre of a Hypersonic Glider in the Presence of
Uncertainties
Sanchito Banerjee1
University of Queensland, Brisbane, School of Mechanical and Mining Engineering, Queensland, 4072, Australia
Zhongjie Wang2, Bernhard Baur3 and Florian Holzapfel4
Technical University Munich, Institute for Flight System Dynamics, Garching, 85748, Germany
A lateral/directional manoeuvre methodology of a hypersonic glider is presented. The
aerodynamics is based on NASAs Generic Hypersonic Aerodynamics Model Example. The
glider weighs 93.1kg with a planform area of 0.78m2

, characteristic length of 0.512m

and a reference length of 1.99m. The control methodology used to carry out a pulsed bank
manoeuvre of

25o deg is presented. A dynamic pole placement controller is implemented

to carry out this manoeuvre. The side slip angle controller is implemented to maintain the
angle at

deg. The baseline controllers are augmented with an

L1

adaptive

controller to cancel out the matched uncertainties. The sources of uncertainties are reduced
control surface effectiveness, aerodynamic uncertainties and control surface failure. The
lateral dynamics are considered in this paper. While the vehicle is descending, a pulsed bank
manoeuvres carried out with the help of the roll angle controller and the side slip angle
controller is presented. Both the pole placement controllers for the lateral/directional
dynamics are Single Input-Single Output (SISO) systems. The fundamental difference
between the baseline and the augmented dynamic pole placement controller with the

L1

adaptive augmentation are presented in terms of robust stability and robust performance.
With the help of 25-run Monte Carlo simulation the performance and the robustness of the
baseline controller are presented. The baseline roll angle controller and the baseline side slip
angle controller show good tracking. The baseline controller exhibits degradation in the
performance in the presence of uncertainties. The augmentation configuration is able to
improve the performance of the baseline controllers significantly. This is outlined with the
help of the tracking error norm,

e pL

Nomenclature

I 11 , I 22 , I 33 =

roll, pitch and yaw principle moments of inertia,

kg m2

1 PhD Student, Division of Mechanical and Mining Engineering.


2 PhD Candidate, Institute for Flight System Dynamics, Technical University Munich.
3 PhD Candidate, Institute for Flight System Dynamics, Technical University Munich.
4 Professor, Institute for Flight System Dynamics, Technical University Munich, Senior Member AIAA.
1
American Institute of Aeronautics and Astronautics

I 13
A
Am
I

aB

= component x-z of moment of inertia matrix,

kg m2

= system matrix
= closed loop system matrix
= body acceleration with respect to inertial frame

= input gain matrix

= output matrix

= wing chord length, m

= gravity vector

I yy

= pitch moment of inertia, kgm2

kx

= autopilot feedback gains

= rolling moment, Nm

= lift coefficient slope as a function of angle of attack, m/s2

Le

= dimensional lift slop with elevator deflection derivative, m/s2

= Dutch roll derivative

Lp

= roll damping derivative

La

= roll control derivative

= mass of vehicle, kg

= pitching moment, Nm

= dimensional longitudinal stability, 1/s2

Mq

= dimensional pitch damping, 1/s2

M e

= pitching moment control derivative, 1/s2

= dimensional normal force slope derivative, m/s2

N e

= normal force slope with elevator deflection derivative, m/s2

= beta yawing moment derivative

Np

= roll rate yawing moment

Nr

= yaw damping derivative


2
American Institute of Aeronautics and Astronautics

= yaw rate, Nm

= roll rate, rad/s

= pitch rate, rad/s

= rate of change of pitch rate

= yaw rate, deg/sec

= velocity vector

= state vector

^x

= state estimate vector

^y

= output estimate

= beta side force derivative

Y a

= Side force derivative as a function of aileron deflection

Y r

= Side force derivative as a function of rudder deflection

= side slip angle, deg

= elevon deflection, deg

= elevator deflection, deg

r
c

= rudder deflection, deg


x

= error in aerodynamic coefficients


= damping ratio

Abbreviations
DPPC
GHAME
LTI
LTV
MIMO
SISO

= Dynamics Pole Placement Controller


= Generic Hypersonic Aerodynamics Model Example
= Linear Time Invariant
= Linear Time Varying
= Multi Input Multi Output
= Single Input Single Output

I.

Introduction

ypersonic flight control is a challenging research area. This is due to the fact that the controller needs to
deal with large flight envelope, rapidly changing aerodynamic coefficients and the uncertainties associated
with it, coupling between the structure and the aerodynamics of the vehicle [8,9,11]. In order to ensure
control authority over the entire flight envelope, control laws need to be developed that are robust to uncertainties
and are able to deal with large parameter variations.
3
American Institute of Aeronautics and Astronautics

There is very limited study presented on the lateral dynamics of a hypersonic glider. The state space
representation of the lateral dynamics of the X-29 is placed in [14]. However, this representation is the coupled
dynamics of the vehicle. The baseline controller presented in this paper is a LQR controller. This setup is augmented

L1 adaptive controller. The augmentation is setup in a way to cancel the uncertainties in the model. The

with a

paper presents results of the performance of the augmentation in the presence of unmatched uncertainties and cross
coupling effects. In the presence of a failure the augmentation setup is able to restore nominal performance. [12] and
[13] outline the static stability characteristics of a hypersonic glider for different configurations. The stability
characteristics of a flat top configuration are outlined in [12] for Mach numbers that are being considered in this
paper. The range of Mach numbers being considered in this paper is Mach 0.6 to 8. There are several papers that
cover the adaptive control of the longitudinal dynamics of hypersonic vehicles [3-7]. The same methodology is used
in this paper to formulate the control problem and carry out the controller design.
This paper is organized as follows: Section II introduces the model of the hypersonic vehicle. In Section III, the
control methodologies are presented. This section also develops the controller design and presents the sensor that is
used in the simulations. The performance, stability and robustness analysis is placed in Section IV, while the final
remarks are given in Section V. The main contributions of this paper are as follows:

Comparison of the baseline controller to the

L1 augmented controller for a LTV system to control

the lateral/directional dynamics of a hypersonic glider.

The derivation and presentation of the theoretical performance and stability bounds of the

L1

adaptive augmented controller.

The presentation of the nonlinear adaptive augmented control law that can be utilised for nonminimum
phase state feedback LTV systems.

This is carried out in the presence of reduced elevon effectiveness, rudder failure, cross coupling effect and
aerodynamic and gravimetric uncertainties.

II.

Model of Hypersonic Vehicle

The hypersonic glider used in the simulations is a scaled down version of the Generic Hypersonic Aerodynamics
Model Example (GHAME) model presented in [1](pp 404-407). The aerodynamic coefficients of the glider are
taken from NASAs GHAME model. This model is used as it provides a complete list of coefficients in the form of
lookup tables. The values of the coefficient range over the flight conditions with Mach numbers are

M [0.4 24 ] , and angle of attack [ 3 21 ] 0 . The physical properties of the vehicle are placed in Table 1
below. The glider gravimetrics are taken from [7].
Table 1. Glider Model Gravimetrics
Mass
(kg)
93.1

Planform
Area, S
(m2)
0.78

Characteristic
Length (m)
0.512

Reference
Length
(m)
1.99

I11 (kg.m2)

I22 (kg.m2)

I33 (kg.m2)

I13 (kg.m2)

1.82

30

32.25

-2.41

Linearised plant models that are used to describe the lateral dynamics are presented. The lateral dynamics of a
hypersonic glider can be represented as a multiple input multiple output (MIMO) system. This system is placed in
Eq. 1 and is taken from [18].

4
American Institute of Aeronautics and Astronautics

[][

N
Y
V
L
0

is

again

Nr
r
1
=
p
Lp

This

representation

once

x (t)= A (t)x ( t )+ B(t )u( t) . Where,

Np
0
Lp
1

0
g
V
0
0

in

A R

][ ] [

the

n n

N a
r
Y a

+ V
p
La

N r
Y r
a
V r
L r
0

common

B R

and

[ ]

state

nm

(1)

space

representation

form

. The states of the system are yaw rate,

sideslip angle, roll rate and roll angle.


Once again the behaviour of the glider dynamics is considered without any control input. This is called the
homogeneous system. The Matrix,

A , has 4 eigenvalues. Typically the lateral dynamics has 2 real eigenvalues

and a complex conjugate pair. The following different motions are observed in the lateral dynamics:

Roll dynamics (R): The rotation about the roll axis. Its aperiodic, relative quick and is stable.

Spiral Mode (S): the spiral motion is relatively slow and can also be unstable.

Dutch Roll (D): The Dutch roll dynamics is depicted with the help of conjugate pair.

The characteristic equation for the lateral dynamics is

( s+ T1 )( s+ T1 ) ( s +2 s + )=0
2

2
D

(2)

In order to simply and decouple the system, the coupling effects of the roll dynamics
dynamics

(r , )

Np

are not considered. As a result

and

g
V0

( p , ) on the yaw

are defined as zero in the model.

Consequently the state space representation of the lateral dynamics is in the form:

x =

A 11 0
x (t ) + B(t) u(t)
A 21 A 22

(3)

Now the dynamics of the entire system, if coupling is not taken into account, can be described with the help of
the diagonal sub matrices represented through the

A 11 and the A 22 matrices.

The roll dynamics of the system can be represented using the

A 22 matrix. The simplified representation of

the roll dynamics leads to


5
American Institute of Aeronautics and Astronautics

[ ] [ ][ ] [ ]

p
L p 0 p L
= 1 0 + 0 a
a

(4)

The characteristic equation for this system is:

det

Comparing the above characteristic to

([

])

sL p 0
=s ( sL p ) =0
1
s

s s+

1
TR

(5)

the time constant for the roll dynamics is

T R=

1
Lp .

The spiral mode degrades due to the integrator and also due to the simplification of the system.

A 11 , describes the Dutch roll dynamics. The state space

The northwest section of Eq. 1, Matrix

representation of the Dutch roll dynamics is as follows

][ ] [ ]

[][

N
N
r +
Y
Y r

V
V

Nr
r
= 1

(6)

The characteristic equation for this subsystem is

det

([

sN r
1

])

N
Y
+ N =0
Y =( sN r ) s
V
s
V

( s2 +2 s+ 2 )

Comparing the coefficients with the standard form

D = N +
D=

leads to

Y Nr
N
V

Y
1
N r
2 D
V

(8)

(9)

The natural frequency for the Dutch roll dynamics is determined through the derivative
for stability

N >0 and therefore consequently follows that

(7)

Cn >0

N . This means that

. It can therefore be concluded that the

6
American Institute of Aeronautics and Astronautics

vehicle is statically stable about the lateral axis. This means that due to a disturbance if there is a sideslip angle, a
counter moment is generated to restore the value to its trim value.
Linearised plant models are also used to describe the longitudinal dynamics which are used to control the altitude
of the glider. The flight path angle controllers are taken from [1](pp 411-422). The plant model is used to carry out
the gain calculations for pole placement controller. The most prominent aerodynamic and control dimensional
derivatives of the model are the pitch damping,
dependency on the angle of attack,

M q , static longitudinal stability,

and the control moment derivative,

M , normal force

M e . The relationship between

the non-dimensional and the dimensional derivatives are placed in [1](pp 411). The non-dimensional derivatives are
stored in look-up tables and are converted to the dimensional values in real time during flight.
The flight path angle dynamics are derived in [1] (pp421-422). The flight path angle state space model is
represented in Eq. (10). Flight path angle tracking is used to control the altitude of the glider.

[][

][ ] [ ]

M q M M
M e
q
q
1
0
0
+ 0 e

=
L L
Le

0
V
V
V

with the dimensional derivatives

L ,

Mq ,

M ,

Mq ,

(10)

M e .In this paper the performance of the

flight path angle controller is not presented as the performance and the robustness of this controller is placed in [15].
Vehicle Plant Dynamics
The actual closed loop plant dynamics of the vehicle is placed in Eq. 11.

x ( t )= A M ( t ) x ( t )+ B1 ( t ) K r r ( t )+ B1 ( t ) ( u a ( t ) + 1 ( t ) ) + B2 ( t ) 2 ( t ) , x ( 0 )=x 0
y=cT x ( t )

where

AM

closed loop plant model which includes the baseline controller.

B 2 is chosen so that
unmatched uncertainties.
uncertainties.

(11)

K r

B T1 B 2=0 . B1 and
1
1
m

B 2 are the control gain matrices for the matched and the

includes the matched uncertainties and

1
c A ( t ) B1 ( t )
T

B 1 is the control input matrix.

is the feedforward gain.

III.

2 includes the unmatched

ua is the adaptive control signal.

Controller Design

This section presents the control methodologies that have been employed to carry out the study. The baseline
controller is implemented with the help of a pole placement technique. The second controller that is implemented for
7
American Institute of Aeronautics and Astronautics

this study is the

L1 adaptive controller as outlined in [2] (pp 211-223). A more detailed look at these control

strategies are placed in the Section A and B. The augmented setup of the pole placement controller and the

L1

adaptive controller are presented in Fig. 3.

Fig. 1 Controller Augmentation Configuration


A. Baseline Controller
This section of the report outlines the derivation of the pole placement controllers. A general state space model
has the following form as in Eq. 12:

y (t )=h T x (t )

x ( t )=F ( t ) x ( t )+ g ( t ) u ( t )
F (t ) R

n n

is the system matrix,

the states of the vehicle,

g (t) R

n1

u ( t ) is the control input and

is the control input matrix,

(12)

x (t ) R

n 1

contains

y (t) is the output of the system.

The block diagram of controllers is placed in Fig 2.

[]
x1
x2

Fig. 2: Structure of Yaw and Roll Controller


The gains of the pole placement controller for the yaw and roll controllers are determined using the Eq. 13
presented in [1]:

8
American Institute of Aeronautics and Astronautics

Det ( IsF ( t )+ g ( t ) c )= ( s pi )

(13)

i=1

Where

pi are the desired pole locations and c

is the feedback gain matrix. Both the systems are of the

second order the desired dynamics are represented using the second order equations:

s 2 +2 n s +2n

Where

represents the desired damping ratio of the system and

(14)

n is the desired natural frequency.

Substituting the state space models for the roll and the yaw controllers, the feedback gains are presented below. For
the case of the glider as the conditions are changing rapidly the desired characteristics are a function of a trajectory
parameter; in this case the dynamic pressure. The non-scheduling of the desired response leads to significantly
reduced performance. The performance of the unscheduled baseline controller corresponds to test case S1 and S2.
As a result, the gains for the side slip controller are as follows:

kr =

Y + N r V VY k + 2V n
r

N V
r

V
k =

(15)

N Y r ( Y + N r V +2 V )
Y ( Y + N r V +2 V )
N r
N V + Nr Y
+
2
V
V
N V Y 2r
N r V Y Y r + N r V Y r +
N r

(16)

And the controller gains of the roll controller are as follows:

k p=

2 n+ L p
L
a

(17)

2n
L

(18)

k =

The feed-forward gain for both the controllers is placed in Eq.19 below:

K ff

1
c A (t ) B (t ) .
T

1
m

9
American Institute of Aeronautics and Astronautics

(19)

A pole placement technique is employed to control the flight path angle dynamics of the vehicle. For maintaining
a certain glide altitude a flight path angle controller is used. The flight path angle controller is shown in Fig 3. This
controller uses pitch rate, pitch angle and flight path angle as the states.

Fig. 3: Flight Path Angle Controller


Substituting the above plant model into Eq. 20 below in order to calculate the feedback gains.

|[

]|

I nxn sF+ g ( c +G P h ) G I
s
Det
=
(s pi )


i=1
h

(20)

Therefore the gain formulas for the flight path angle controller are as follows:

[]

Me

c1
M L
L M
c 2 = e e
V
V
c3
0

where

0
M e
M e L L e M

V
V

is the desired closed loop natural frequency,

Le
V
M q L e
V
M e L Le M

V
V

][ ]
1

is the damping ratio and

L
V
L M
bm + M (21)
+ q
V
cm
am + M q

is one real pole

location is required to calculate the adaptive gains of the controller. The gains are calculated at the controller loop
rate in the simulation or during flight real time. The feedforward gain, as depicted in Fig 3 is derived for the
controller to have perfect tracking as is the same as in Eq. 19. A more detailed performance of the flight path angle
controller is placed in [15].
B.

L1 Adaptive Controller
In this section, the

L1 adaptive controller is introduced. For this study the

to augment the SAPPC. The philosophy behind this

L1

L1 adaptive controller is used

adaptive controller is to obtain a state feedback

controller for the unmatched uncertainties and gets its estimates of these uncertainties and to compensate for these
uncertainties within the bandwidth of the low pass filter. The filter ensures that the controller remains in the low
10
American Institute of Aeronautics and Astronautics

frequency range in the presence of fast adaptation and large reference inputs [2]. This adaptive control scheme is
chosen for the following reasons [2, 11]:

Decoupling of the rate of adaption and robustness achieved with the help of the low pass filter
Guaranteed fast adaption, limited only by hardware constraints.
Guaranteed, bounded away from zero time delay margin
Guaranteed transient performance for a systems input and output signal, without high gain feedback or
enforcing persistent excitation type assumptions.
Reduced cost for the validation and verification process: achieved with the help of the analysis that is
carried out on the

L1 reference system and the L1 design system.

The following few sections outline and present the main components of the

L1

adaptive controller. The

L1 controller setup employed in this paper is placed in Fig 4.

Fig. 4: Architecture of L1-Adaptive Controller


State Predictor:

^x ( t )= A x^ ( t )+ B K r ( t ) + B ( u ( t ) + ^ ( t ) ) +B ^ ( t ) , ^x ( 0 )=x
M
1
r
1 a
1
2 2
0
^y =cT x^ ( t )

where

^ 1 ( t )

K r

and

^ 2 ( t )

(22)

are the estimates of the matched and the unmatched uncertainties respectively and

1
c A ( t ) B1 ( t ) . The estimates of these uncertainties are provided by the piecewise wise constant law.
T

1
m

Control Objective:
11
American Institute of Aeronautics and Astronautics

Ideal desired system is:

x M = A M x M + BM r
y M =C P x M
y M =C P ( s I n A M )1 B M r ( s )

Defining

M ( s ) as transfer function from r to

(23)

y M , which would be:


1

M ( s ) C P ( s I n A M ) B M

(24)

Control Law:

ua ( t ) is the adaptive control law and is presented next. The matching condition in the presence of unmatched
uncertainties is (equating the current system to the control objective):

Y P ( s )=Y M ( s )
H 1 k r r ( s )=H 1 K r r ( s ) + H 1 ( ua + 1 ( s ) ) + H 2 2 ( s )
0=H 1 ( u a+ 1 ( s )) + H 2 2 ( s )
H 1 ( u a + 1 ( s ) )=H 2 2 ( s )

( ua + 1 ( s ) ) =H1
1 H 2 2( s )
ua , ideal=H1
1 H 2 2 ( s ) 1 ( s )

Where

H 1=C ( sI A M )1 B1 and

H 2=C ( sI A M )1 B 2 .

(25)

B1

and

B2

are the control gain

matrices for the matched and the unmatched uncertainties. As the values of the uncertainties are not known, the
estimates of the uncertainties (obtained from the piecewise constant update law) are used to formulate the control
law. Furthermore, as only certain frequencies of this uncertainty can be cancelled out in Eq.25, therefore is a need to
add low pass filter as part of the adaptive control signal. Therefore the adaptive law is:

^( ) ^( )
( )
ua=C ( s ) ( H 1
1 s H 2( s) 2 s 1 s )

12
American Institute of Aeronautics and Astronautics

(26)

Where

C( s)

is a vector of low pass filters. The two filters

C1 ( s )C 2 (s )

are for the matched and

unmatched uncertainties respectively. Due to the time varying nature of the plant and the control input matrix

H 1
1 ( s ) H 2 (s )
is scheduled as a function of dynamic pressure and Mach number.
Piecewise Constant Adaptive Law:
The piecewise constant law that defines the estimated values of the uncertainties is placed in Eq.27 below.

^ ( iT S )=B1 1 ( T S ) ( T S ) , t

(27)

where:

^ ( iT S )=

[ ][

^1 ( i T S )
= matched uncertainties
^2 ( i T S ) unmatched uncertainties

A
( T S ) A1
M (e

TS

e p=~
xP = ^x x p

I n )

( T S )=e

AMT S

e P ( iT S )

B=[ B1 B2 ]

T S=Sampling period of the model


The piecewise constant adaptive law can be rewritten from an implementation point of view. The following
component of the adaptive control law is constant in Eq. 28:

^ ( iT S )=(B1 1 ( T S ) e A

TS

) e P ( iT S )

(28)

Low Pass Filter:


The structure of the low pass filter is give in Eq.29.

C ( s)

b
s + b

(29)

b is the bandwidth of the filter. This simple structure is used to filter the adaptive signal in the matched and
unmatched channel.
Stability and Performance Bounds
Reference system with the inclusion of the reference adaptive control signal is:
13
American Institute of Aeronautics and Astronautics

x ref = A M x ref + B1 K r r (t)+B 1 [ uref + f 1 ( xref , t ) ]+ B 2 f 2 ( x ref ,t )


y ref =C P x ref , x ref ( 0 )=x 0
AM

where

Kr

(30)

includes the state feedback gains of the system. This matrix describes the closed loop system.

B 1 is the input gain matrix and

is the feedforward gain.

B 2 is its null space.

uref

here depicts

C P is the output matrix.

the adaptive signal from the augmentation.

Rewriting the reference system in frequency domain leads to the following representation:

x r ef ( s )= ( s I n A M )1 [ B1 ( u ref ( s ) + 1,ref ( s ) + K r r (t) ) + B2 ( s ) 2, ref ( s ) + x 0 ]


y ref ( s )=C P xref ( s )
i ,ref (s)

where

(31)

is the Laplace transform of:

whose Laplace transform is

i ,ref ( s )=L { f i ( X ref , t ) } ,i=1,2

and defining

x (t) ,

x ( s ) ( s I n A M )1 x 0 , to depict the initial conditions. Therefore:

x ref ( s )= ( s I n A M )1 [ B1 ( u ref ( s ) + 1,ref ( s ) + K r r (s) ) + B 2 ( s ) 2,ref ( s ) ] + x ( s (32)


)
y ref ( s )=C P ( s I n A M )1 [ B1 ( uref ( s ) + 1,ref ( s ) + K r r (s) ) + B2 ( s ) 2,ref ( s ) ] +C(33)
P x ( s)
To make the above formulation more readable the following intermediate variables are introduced:
1

H x B ( s ) ( s I n A M ) B 1
1

H x B ( s ) ( s I n A M ) B2
2

H 1 ( s ) C P H x B ( s ) =C P ( s I n A M )1 B 1
1

H 2 ( s ) C P H x B ( s )=C P ( s I n A M )1 B 2
2

Using these intermediate variables leads to:

x ref ( s )=H x B ( s ) [ ( uref ( s )+ 1,ref ( s ) + K r r (s) ) ] + H x B ( s ) 2,ref ( s )+ x ( s )

(34)

y ref ( s )=H 1 ( s ) [ uref ( s ) +1,ref ( s ) + K r r (s) ]+ H 2 ( s ) 2,ref ( s )+ C P x ( s )

(35)

As

B1

is known, a prefilter

K Rm m

can be chosen to render

B 1 K r =BM , which leads to

M ( s ) =C P ( s I n A M )1 B M =C P ( s I n A M )1 B1 K r . Therefore, the ideal reference output is as follows:


y , id ( s ) = y M ( s )=M ( s ) r ( s ) =C P ( s I n A M )1 B1 K r r ( s )=H 1 ( s ) K r r ( s )
The ideal reference control signal

uref ,id ( s) is derived next. In order to derive the formula for

(36)

uref ( s) the

influence of the initial condition is omitted. As previously mentioned, it can be shown that non-zero initial condition
only leads to an exponentially decaying additional contribution. This leads to:
14
American Institute of Aeronautics and Astronautics

y ref ( s )=H 1 ( s ) [ uref ( s ) +1,ref ( s ) + K r r (s) ]+ H 2 ( s ) 2,ref ( s ) y , id ( s )=H 1 ( s ) K r r ( s )


H 1 ( s ) [ uref ( s ) +1,ref ( s ) ]=H 2 ( s ) 2, ref ( s )
uref ( s )+ 1,ref ( s )=H1
1 ( s ) H 2 ( s ) 2,ref ( s )
uref ,id ( s )= 1 [ H1
1 ( s ) H 2 ( s ) 2, ref ( s ) +1, ref ( s ) ]

(37)

As mentioned previously, it is desirable to avoid high frequency content in the control channel, which is key to

L1

C ( s)

control theory. Therefore, a bandwidth limiting low-pass filter

is added to the control signal

uref ( s) for the reference system:


uref ( s )= 1 C ( s ) [ H 1
1 ( s ) H 2 ( s ) 2,ref ( s )+ 1,ref ( s ) ]

(38)

Using the inverse DC gain method, the modified control law is as follows:

uref ( s )= 1 C ( s )

1
( s ) H 2 ( s ) 2,ref ( s ) + 1,ref ( s )
1
C p A M B 1

Substituting the above control signal back into the plant dynamics leads to the closed loop

[ ((

x ref ( s )= ( s I n A M )1 B1 1 C ( s )

(( [

H x B ( s ) C ( s)
1

(39)

L1 reference system.

])

1
( s ) H 2 ( s ) 2, ref ( s )1,ref ( s ) + 1,ref ( s ) + K r r (s) + H x B ( s ) 2,ref ( s ) + x ( s )
C p A1
M B1

H x B ( s ) ( I mC ( s ) ) 1, ref ( s ) + H x B ( s ) C ( s )
1

1
( s ) H 2 ( s ) 2,ref ( s ) + H x B ( s ) 2,ref ( s ) + x ( s )+ H x B ( s
1
C p A M B1

H x B ( s ) ( I mC ( s ) ) 1, ref ( s ) + H x B ( s ) C ( s )

Where:
1

GB = H x B ( s ) C ( s )
P

1
( s ) H 2 ( s )+ H x B ( s )
1
C p A M B1
2

The closed loop reference system is:

x ref ( s )=G B ( s ) 1,ref ( s ) +G B ( s ) 2, ref ( s ) + x ( s ) + H x B ( s ) K r r ( s )


P

1
( s ) H 2 ( s ) + H x B ( s ) 2,ref ( s )+ x ( s ) + H x B ( s ) K r r
C p A1
M B1

G B =H x B ( s ) ( I m C ( s ) )
P

])

1
( s ) H 2 ( s ) 2,ref ( s ) +1,ref ( s ) + 1,ref ( s )+ K r r ( s) +
C p A1
M B1

y ref ( s )=C p x ref (s )

15
American Institute of Aeronautics and Astronautics

(40)

According to the small gain theorem the following condition apply when calculating the expression for the

L1

norm condition.

x ref L G B L 1, f L +GB L 2, f L +H x B K rL r L +x L
P

x ref L

To determine the bound of

GB L
P

GB L
P

r L

and

For the bound of

x L

are determined using system matrices

r (t) . So the following bounds need to be determined:

and

(41)

every term on the right hand side of the equation needs to be known.

H x B K rL
1

1, f L

2, f L

and

C ( s) , K r

x L

the following relationship is sufficient:

x L

(42)

where ( s I n A M )1L 0
1

where 0 x 0L

Using the definition of the truncated norm leads to the following definition:

x L x L r L rL

(43)

Therefore, now the small gain theorem is written as:

x ref L G B L 1, f L +GB L 2, f L +H x B K rL rL +
P

The next step is to derive the bounds for

(44)

i , f L , where i=1,2

Certain assumptions need to be made before the bounds of


1.

i , f L

is calculated. There are as follows:

The internal dynamics are stable with respect to the states of the reference dynamics. The assumption is
such:

LZ , B Z >0 such that for all t 0, y I L LZ (x ref L + 1 )+ B Z

Where

2.

+
1 R which is derived later, and B Z is an upper bound of

(45)

yI .

The reference state is bounded as:

x ref L < r

(46)

Following these two assumptions:

y IL < LZ ( r + 1 ) + BZ

16
American Institute of Aeronautics and Astronautics

(47)

X ref [ x ref , y I ]

X ref

Then the bound condition for

is as follows:

X ref L < max { r , LZ ( r + 1 ) + B Z }

(48)

Defining an intermediate variable leads to:

r ( r ) max { r , LZ ( r + 1 ) + B Z }
Where

(49)

+
1 1 R is the upper bound of 1

Therefore:

X ref L < r ( r )

Along with the plant assumptions which are:

f i ( t ,0 ) < Bi 0 ,i =1,2

1.

2.

>0, K 1 , K 2 , such that:

for arbitrary

f i ( X 1 , t )f i ( X 2 ,t ) K i X 1 ( t ) X 2 ( t ) ,i=1,2

for all

X j (t ) , j=1,2,

t .

uniformly in

This leads to:

i , ref L K i ( ) X ref L + Bi 0 < K i ( ) r ( r ) + Bi 0 ,i=1,2


r

i.e.

[ ( t , X

Where

ref

(50)

)i ( t , 0 ) ]+ i ( t , 0 )L r ( r )[ X ref ( t )0 ]L + Bi 0

0 X ref < r ( r )

Further defining: Li

[K

i r ( r )

r ( r ) ]

, for every

r >0

Then the last equality is written as:

i , ref L < Li

r + Bi 0 ,i=1,2

(51)

Therefore substituting it into Eq. Error: Reference source not found leads to:

x ref L G B L ( L1

r +B 10) +GB L ( L2 r + B20 ) +H x B K rL rL +


P

Redefining variables as follows:

17
American Institute of Aeronautics and Astronautics

(52)

l0
The

L2
L1

B20
l0

B 0 max B10 ,

and
r

L1 norm condition is re-written as:

x ref L (GB L +G B L l0 ) ( L1
P

r + B 0) +H x B K rL rL +
1

(53)

r , this leads to:

As the bound of the states of the reference states is

r > (G B L +G B L l 0 ) ( L1 r + B0 ) +H x B K rL rL +
P

(54)

And therefore the norm condition is:

GB L +GB L l0 <
P

r H x B K rL rL
1

( L1

(55)

r +B 0 )

So far the derivations are based on the assumption that the bounds for the states of the reference system are met.
This means that at this stage, this proof has not been provided. The proof for this bound is provided through
contradiction.
Assuming that the
assumption

L1

norm condition and stability of the unmodelled internal dynamics hold, but the

x ref L < r

does not hold.

In this case, since

x ref

is continuous,

1 ( 0, ] , such that:

x ref ( t )< r , t ( 0, 1 ]

x ref ( 1 )=r
x ref L =r

This means that


And therefore

(56)

X ref L max { r , LZ ( r + 1 ) + B Z } r ( r )
1

Similar to the previous derivation:

i , ref L < Li

r + Bi 0 ,i=1,2

(57)

This means that, all the reverse substitutions leads to:

x ref L G B L ( L1
1

r + B10 ) +GB L ( L2 r + B20 ) + H x B K rL rL +


P

(58)

As the norm condition hold, the terms on the right hand side of the equation on the whole smaller than
which leads to
all

x ref < r
1

. And this is a contradiction. Therefore, this means that

t [ 0, ] .

The next step is to provide the bounds for

uref L

18
American Institute of Aeronautics and Astronautics

x ref L < r

r ,

is true for

uref ( s ) is as follows:

The formula for

uref ( s )= 1 C ( s )

1
( s ) H 2 ( s ) 2,ref ( s ) +C ( s ) 1, ref ( s )
C p A1
M B1

(59)

This leads to:

uref L 1 C ( s )L

i , ref L < Li

Recalling that

, this leads to:

1
H2
C p A1
B
M
1

ur ef L < u

Uniform bound of

e PL

1
H2
1
C p AM B1

r +Bi 0

u 1 C ( s )L

This leads to:

2,ref L +1,ref L

L1

(60)

(61)

L1 r + B10 ) + ( L2 r + B20 )
r

L1

For the analysis of the transient and steady state performance, there are a few assumptions that need to be made.

There is a time

such that:

x P L with =r + 1

(62)

uL u u=u + 2

(63)

Where:

+
r R is the upper bound of x ref
+
u R is the upper bound of uref
r

1 is the upper bound of 1

uref uL

2 is the positive constant, representing the upper bound of

The 3rd assumption of the uncontrolled plant dynamics leads to:

LZ , B Z >0 , such that for all t 0, y I L L Z x P tL +B Z


t

That:

y I L LZ +B Z

And this leads to:

X P L

max { , LZ + BZ } = r ( r )

Using the above inequality, together with the boundedneness of

f i ( t , 0 ) as given in 1st problem assumption, the

semi-global Lipschitz condition in 2nd problem assumption and the redefinition of

19
American Institute of Aeronautics and Astronautics

Li

, it is concluded that:

i L K i X L + Bi 0 K i

r ( r ) + Bi 0 =Li + Bi 0
r

(64)

1 ( t )2 m1 L =1 L m ( L1 + B10) m

(65)

2 ( t )2 ( L2 +B 20) nm

(66)

r ( r)

r ( r )

This implies that


2

And similarly:
r

LP spaces: Relationship between

Intermediate Theorem of

Cn where 0<r < p :

In the general case, for vectors in

Using the above relationships along with

p norms

(1r 1p )

xp xr n

xp

e I ( ( i+1 ) T S ) = 2 ( ( i+1 ) T S )

which is obtained at the end of the

u1 +u2p u1p+u2p

, the bound for the prediction error is

adaptation law and the Minkowski inequality

given at an instant of time as a function of sampling time:

TS

TS

e ( ( i+1 ) T ) e
P

TS

AS ( T S )

[ BP ( 0 ) u ( i T S+ ) +1 ( iT S + ) ] d

TS

AS ( T S )

BP 1 d +
2

TS

AS ( T S )

BP 2 d =
2

A S (T S )

[ BP 2 ( iT S + ) ] d

TS

A S (T S )

A S (T S )

B P d 1 + (67)
e
2

)m

2 ( L2 + B20 ) nm
r

( T S ) 1 ( T S ) 1+ 2 ( T S ) 2
Where:
TS

1 ( T S ) e

A S (T S )

B P d

A S (T S )

B P d

TS

2 ( T S ) e
0

Having an upper bound for arbitrary


bound that holds for all

( i+1 ) T S , one can rewrite

e ( ( i+1 ) T ) ( T
P

to obtain a

iT s :

e P ( i T S )2 ( T S ) , i T s

(68)

It is important now to establish a bound that concludes that the prediction error remains bounded between sampling
instances.
20
American Institute of Aeronautics and Astronautics

B P d 2 ( T S

Where

1 max {( 0 ) } u + L1 +B 10

e P ( iT S +t ) =e

ASt

e P ( i T S ) + e

t
AS ( t )

B ^ ( iT S ) d e A

( t )

0 ) u ( iT S + ) +1 ( iT S + )
[ B P ( (69)

Introducing the following terms:

1 ( t ) e A t2
S

2 ( t ) e A ( t ) 1 ( T S ) e A
S

0
t

3 ( t ) e A

( t )

B P2 d

( t )

B P2 d

TS

0
t

4 ( t ) e A
0

The following conditions apply to the above terms

i ( T S ) =max i (t)wherei=1,2,3,4

(70)

Substituting the above terms back into the prediction error equation leads to:

e ( ( i+1 ) T ) ( t )e
P

( i T S )2 + 2 ( t )e P ( iT S )2 + 3 ( t ) 1 + 4 ( t ) 2

( 1 ( t )+ 2 ( t ) ) ( T S )+ 3 ( t ) 1+ 4 ( t ) 2 t [ 0, )
Replacing the term

(71)

i ( t ) with its maximum values leads to:

e P ( T S )2 ( 1 ( t ) + 2 ( t ) ) ( T S ) + 3 ( t ) 1 + 4 ( t ) 2 t [ 0, )

(72)

Introducing another intermediate variable

0 ( T S ) ( 1 ( t ) + 2 ( t ) ) ( T S ) + 3 ( t ) 1 + 4 ( t ) 2
(73)

t [ 0, )
This means that

e P ( t )2 0 ( T S ) t [ 0, )

(74)

e P L 0 ( T S )

(75)

This implies that:

This means that the prediction error can be rendered arbitrary small with the choice of

T S . The limiting

relationship that expresses this statement is:

lim 0 ( T S )=0

(76)

TS 0

Therefore, it is important to choose

T S in a way so that:

0 ( T S ) < 0

(77)
21
American Institute of Aeronautics and Astronautics

This in turn means:

e P L < 0

(78)

To prove the property that

lim 0 ( T S )=0

TS 0

, it is sufficient to show that:

lim ( T S )=0

(79)

TS 0

And

lim 3 ( T S )=0

TS 0

From the definition of

lim 4 ( T S ) =0

and

TS 0

, since

1(T S ) , 1 ( T S ) ,

1 ,

2 are bounded.

1 ( T S ) and 2 ( T S ) , it is concluded that:

lim 1 ( T S )=0 lim 2 ( T S )=0

TS 0

(80)

TS 0

i are continuous, it leads to:

Since all the signals pertaining to

lim 1 ( T S ) =0 , lim 2 ( T S )=0 , lim 3 ( T S )=0 , lim 4 ( T S ) =0

TS 0

TS 0

T S 0

T S 0

(81)

With the help of these relations, the bound for the prediction error is established.
Performance Bounds
The performance bounds are once again proved with the help of contradiction
It is assumed that the bounds

And

x ref x PL < 1

and

uref uL < 2

do not hold. Then:

x ref ( 0 )x P ( 0 )=0 < 1

(82)

uref ( 0 )u P ( 0 )=0< 2

(83)

x ref , x P , uref u are continuous, there exists

such that:

x ref ( )x P ( )=1

(84)

uref ( ) uP ( )= 2

(85)

While simultaneously:

x ref ( t )x P ( t ) < 1 t [ 0, )

(86)

uref ( t )u P ( t ) < 2 t [ 0, )

(87)

This means that at least one of the inequalities hold:

x ref x P L = 1

(88)

uref uP L = 2

(89)

If one of the equalities does not hold, then it holds that its

norm is less than the respective bound

i , i=1,2
22
American Institute of Aeronautics and Astronautics

x ref ( )x P ( )= 1

Using the reverse triangle inequality and

, it is concluded:

x P L x ref L +x ref x PL = xref L + 1

Or respectively with

uref uP L = 2

(90)

, it is concluded:

u PL uref L +u ref u PL =u ref L + 2

While an unfulfilled equality

x ref x P L = 1

uref uP L = 2

or

results in the respective expression

sign. With the above conditions along with the stability of the internal

with a strict inequality < instead of a


dynamics, the uniform bounds for

(91)

x P L

and

u PL

are satisfied. Furthermore, using the above bounds,

the transient and steady state performance can be satisfied:

e P L < 0

(92)

Next it is important to derive the bound regarding the tracking of the reference state by the system state. For this
purpose, rewriting the state of the plant dynamics in the frequency domain as done with the

L1 reference system

before.
With the definition of

H xBp , H x Bp , x ( s )i (s) , the dynamics are as follows:

x P ( s )=H x B ( s ) ( u a ( s ) + 1 ( s ) ) + H x B ( s ) 2 ( s ) + H x B ( s ) K r r ( s )+ x ( s )
1

The filter is defined as:

Ci ( s )=

(93)

i
, i=1,2
s + i

(94)

The control law is:

u ( s )=C ( s ) ^ 1 ( s ) +

1
c A1
M B1

H B ( s ) ^ 2 ( s )
P

(95)

Expanding this control law one gets:

u ( s )=C ( s )

C (s )

C (s )

C (s )

[
[
[

1
1
M

c A B1

1
1
M

c A B1
1
1
M

c A B1
1
1
M

c A B1

] [

H B ( s ) ^ 2 ( s )^ 1 ( s ) +C ( s )
P

H B ( s ) 2 ( s ) 1 ( s ) +C ( s )
P

]
] [

H B ( s ) 2 ( s )1 ( s ) +C ( s )
P

H B ( s ) 2 ( s )1 ( s ) +C ( s )
P

1
1
M

c A B1

1
1
M

c A B1
1

c A1
M B1

H B ( s ) ^ 2 ( s )C ( s )
P

1
c A1
M B1

1
c A1
M B1

c A B1

H B ( s ) ~
2 ( s ) ~
1 ( s)
P

23
American Institute of Aeronautics and Astronautics

(96)

H B ( s
P

H B ( s ) 2 ( s )C ( s ) ^
P

H B ( s ) ( ^ 22 )C ( s ) ( ^ 1 ( s ) 1 ( s ))

1
1
M

] [

H B ( s ) 2 ( s )1 ( s ) C ( s )

Where

1
1
M

c A B1

H B ( s ) ~
2 ( s ) ~
1 ( s ) =~
(s )
P

u ( s )=C ( s )
The redefinition of

1
1

c A M B1

. Therefore the control law is:

H B ( s ) 2 ( s )1 ( s )+ ~ ( s )
P

(97)

~
( s ) is required and this is done with the help of the error dynamics.
e =A e + B ~ +B ~

2 2

(98)

Resulting the in the following equation in the frequency domain:

e P ( s )=( s I n A S )1 ( B1 ~
1 + B 2 ~
2)
Hx ~
1 ( s ) + H x ~
2 ( s )
B1

B2

1
1
~
~
H 1
x C P e P ( s )=H x C P H x 1 ( s ) + H x C P H x 2 ( s )
B1

B1

B1

B1

(99)

B2

From that follows that:

~
( s )=H 1
x C P eP ( s)

(100)

B1

GB ( s) and GB ( s)

Together with the definition of

~
( s ) the dynamics of the system is as follows:

and

x P ( s )=G B ( s ) 1 ( s )+ GB ( s ) 2 ( s ) + x ( s ) + H x B ( s ) K r r ( s )+ H x ( s ) C ( s ) H 1
C P eP ( s )
x (101)
P

B1

B1

And the reference system is:

x ref ( s )=G B ( s ) 1,ref ( s ) +G B ( s ) 2, ref ( s ) + x ( s ) + H x B ( s ) K r r ( s )


P

(102)

Therefore the following expressions are obtained:


1
x ref ( s ) x P ( s )=GB ( s ) 1,ref ( s )GB ( s ) 1 ( s )+ GB ( s ) 2,ref ( s ) G B ( s ) 2 ( s ) + H x ( s ) C ( s ) H x C P e P ( s )
P

B1

B1

( s ) H 1
x ref ( s ) x P ( s )=G B ( s ) ( 1,ref ( s )1 ( s ) ) +G B ( s ) ( 2,ref ( s ) 2 ( s ) ) + H x ( s ) C (103)
x C P ( s ) eP ( s)
P

B1

B1

Incorporating the bounds leads to:

( )
x ref ( s )x P ( s )L GB ( s )L 1,ref ( s ) 1 ( s )L +GB ( s )L 2,ref ( s )(104)
2 s L +H x
P

Now it is important to find the bounds of

1, ref ( s )1 ( s )L

and

2, ref ( s )2 ( s ) L

. Furthermore

e P ( s )L < 0

With

x P L < r + 1 r + 1 =

and the third assumption on the stability of the unmodeled dynamics, it

holds that:

y I L < LZ ( r + 1 ) + B Z

And hence the definition of

r ( r ) , leads to:

24
American Institute of Aeronautics and Astronautics

(105)

B1

( s ) C ( s ) H 1
x

X PL <max { r + 1 , L Z ( r + 1 )+ B Z } r ( r )

where

X P [ xP , yI ]

(106)

L1 reference system, it has been determined that:

From the analysis of the

X ref L < max { r , LZ ( r + 1 ) + B Z } r ( r )

where

X ref [ x ref , y I ]

(107)

These two bounds given for the uncertainties together with the semi-global Lipschitz condition in the problem
assumption 2 that:

f i ( t , X ref ) f i ( t , X P )

i , ref ( s ) i ( s )L =

(108)

And as:

X ref ( s ) X P ( s )=

[ ][ ][

xref
x
x x P
P = ref
yI
yI
0

it further holds that:

i , ref ( s ) i ( s )L K i X ref ( s ) X P ( s )L
r ( r )

(109)

From two existing definitions which are:

r ( r ) max { r , LZ ( r + 1 ) + B Z } and
Li

[K

i r ( r )

It follows that:

r ( r ) ]

, for every

r >0

K i ( ) < Li , i=1,2

(110)

r r

And therefore one can get:

i , ref ( s ) i ( s )L < Li x ref ( s )x P ( s )L

,
(111)

i=1,2
This results in the following expression:

s )x P ( s )L +H x
x ref ( s )x P ( s )L GB ( s )L L1 x ref ( s )x P ( s )L +GB ( s )L L2 xref ((112)

Collecting like terms leads to:

( 1G

BP

( s )L L1 G B ( s ) L L2
1

)x

ref

( s ) x P ( s )L <H x ( s ) C ( s ) H1
( )
x C
P s L 0
(113)

Defining the following intermediate variable leads to the following:

l 0=

L2

L1

25
American Institute of Aeronautics and Astronautics

B1

B1

B1

( s) C ( s

(1L (G
1

From the

BP

( s )L G B ( s )L l 0) x ref ( s )x P ( s )L <H x ( s ) C ( s ) H 1
0
x C P ( s ) L
P

B1

B1

L1 norm condition:
r H x B K rL rL

GB L +GB L l0 <
P

L1

( L1

(GB L +GB L l0 ) <


P

r +B 0 )

H x B K rL rL
r

<1
B0
B0
r +
r +
L1
L1
1

x ref ( s )x P ( s )L <

x B1

Uniform bound of

uref ( s )u ( s )L

( s ) C ( s ) H1
( )
x C P s L 0
B1

(1L (G
1

BP

( s ) L GB ( s )L l 0 )
1

(114)

Calculating the bounds of the control surface deflection follows a similar process to that of the states.

uref ( s )u ( s )= 1 C ( s )

1
1
( s ) H 2 ( s ) 2,ref ( s )+ 1,ref ( s ) + 1 C ( s )
( s ) H 2 ( s ) 2
1
C p A M B1
C p A1
B
M
1

uref ( s )u ( s )= 1 C ( s ) ( 1,ref ( s ) 1 ( s ) ) 1 C ( s )

1
( s ) H 2 ( s ) ( 2,ref ( s ) 2 ( s ) ) 1 C ( s )
1
(115)
C p A M B1

This leads to

uref ( s )u ( s )L 1 C ( s )L 1, ref ( s ) 1 ( s ) L +
1

C (s )

uref ( s )u ( s )L 1 C ( s )L L1 x ref ( s )x P ( s )L +
1

uref ( s )u ( s )L 1 C ( s )L L1 +

1 C ( s )

1
(s ) H2 (s )
C p A1
M B1

1 C ( s )

2, ref ( s ) 2 ( s )

L1

1
( s ) H2 (s )
1
C p A M B1

1
( s) H 2 ( s)
1
C p A M B1

L2

L1

L1

L2 x ref ( s ) x
r

x ref ( s )x P ( s )L + 1 C
(116)

Defining

1 C ( s )L L1 +
1

1 C ( s )

1
( s ) H2 (s )
C p A 1
M B1

L2

L1

uref ( s )u ( s )L 2

Uniform bound of

y ref ( s ) y ( s )L

26
American Institute of Aeronautics and Astronautics

x ref ( s )x P ( s ) L + 1 C ( s ) H 1
x (s )CP (

(117)

B1

y ref ( s ) y P =C P ( x ref ( s ) x P ( s ) )

(118)

Therefore the upper bound is:

y ref ( s ) y ( s )L <C P ( x ref ( s ) x P ( s ) )L

(119)

Stability Proofs
Since all the above proofs have been provided it is shown that:

x PL <

(120)

uL < u

(121)

C. Sources of Uncertainty Control Surface, Aerodynamic and Gravimetric Uncertainties


Control Surface Failure
After re-entry, the state of the system is unknown. Therefore, there is a possibility that the rudder might not be
functioning or the ailerons might have reduced functionality. It is these two scenarios that are utilised to test the
performance and the robustness of the controllers. For the case of the rudder failure, the rudder is stuck at

2.00

and for the case of the elevons reduced functionality; the control surface is only able to deflect to 50% of the
commanded deflection.
Aerodynamic and Gravimetric Uncertainties
This section outlines the implementation of the corruption of the aerodynamic coefficients. For testing the
performance boundaries of the pole placement controller and the pole placement augmented with the

L1

adaptive controller, the uncertainties of the coefficients can be lumped together into a vector as in [19]:

e=[ C , L , N , N , Y , L , N , Y , L
Mq

where

(122)

X are the multiplicative uncertainties of the aerodynamic and control derivatives. The nominal values of

the error vector is given by


T

e=[ 11 11 1 11 11 ]

(123)

Each element in the error vector is set to a minimum and maximum value. These values are placed in Table 1
below. The distribution that is presented in Table 4 is a uniform distribution and the bounds that are presented are the
maximum and minimum values of the errors. The error bounds are defined based on the guidelines presented in
[16].
Table 2. Errors Bounds of Error Vector (Time Invariant)
Component of Error
Vector

Error Bounds

[0.80,1.20]

Mq

27
American Institute of Aeronautics and Astronautics

[0.70,1.30]
[0.70,1.30]

[0.70,1.30]
[0.70,1.30]

[0.70,1.30]
[0.95,1.05]

[0.95,1.05]
[0.95,1.05]

Values from Table 2 will be incremented and the boundaries in terms of stability and performance of the
controllers will be established. These errors are time invariant errors which are defined at the start of every
simulation and remain constant for the whole simulation. The error margin is defined at the start of the simulation
and the error margin will remain constant for the entire trajectory. 15% uncertainties were introduced to the moment
and product of inertias of the vehicle and also a 5% uncertainty was introduced to the actual mass of the system.
After re-entry there, uncertainties in the gravimetric properties of the vehicle are present. Therefore, it is important
to check for the robustness properties of the controller in the presence of these uncertainties as well.
Combination of Errors
None of the aforementioned uncertainties occurs one at a time during flight. Therefore the effects of a
combination of errors and the controllers robustness to these uncertainties are tested. The following errors are
present in the simulation at the same time:

Moment of Inertia Uncertainties

Mass Uncertainty

Aerodynamic Uncertainties

Rudder Failure-leading to the lack of tracking ability in the

Reduced deflection of the elevons by 50%.

-channel.

The results presented for this scenario is obtained with the help of a 25-run Monte Carlo Simulation.

IV.

Simulation Results and Analysis

This section presents the scenarios that are considered for the Monte Carlo runs. This section further outlines the
variables that are considered for the simulations and the practical implications of choosing these scenarios. The
hypersonic glider is simulated in a high-fidelity 6-DOF simulation in CADAC++. In this platform the sensors errors
are considered during the Monte Carlo runs. The aerodynamic coefficients, which are a function of the states of the
28
American Institute of Aeronautics and Astronautics

glider, are used in the gain calculations for the baseline controller; therefore this presents a proving ground for the
performance and stability boundaries of the baseline controller [11]. The Monte-Carlo analysis that will be presented
in this paper will consist of applying the uncertainties to the simulation vehicle models.
A Simulation Scenarios
In order to evaluate and establish the robustness of the pole placement and

L1

adaptive controllers, the

manoeuvres performed using several different scenarios. It is important to note that the implementation of the
controllers is not changed for the different scenarios. The scenarios are stated in Table 3:
Table 3. Simulation Scenarios
Nominal Conditions: perfect state feedback. Baseline controller

L1

Pole Placement Controller with rudder failure

L2

Pole Placement Controller with reduced aileron functionality

L3

Pole Placement Controller with gravimetric errors

L4

Pole Placement Controller in the presence of aerodynamic uncertainties

L5

Pole Placement Controller in the presence of combination of L2-L5.

L6

L1 Augmented Controller under nominal conditions

L7

L1 Augmented Controller with rudder failure

L8

L1 Augmented Controller with reduced aileron functionality

L9

L1 Augmented Controller with Gravimetric Errors

L10

L1 Augmented Controller in the presence of aerodynamic uncertainties

L11

L1 Augmented Controller in the presence of combination of L2-L5

L12

The simulations in the presence of aerodynamic uncertainties and gravimetric uncertainties are carried out with
the help of a 25-run Monte Carlo run. Cases S3-S12 is carried out with the scheduled baseline controller.
D. Controller Performance and Robustness Analysis

29
American Institute of Aeronautics and Astronautics

This sub-section of the paper includes the preliminary results that have been carried out to this point of the study.
This sections presents the simulation results for a pulsed bank manoeuvre of

250 deg for perfect state feedback.

Table 4 below contains the initial conditions for the trajectory control simulation case.
Table 4. Initial Conditions - Trajectory Control
Altitude
(km)

Pitch Angle
(deg)

Angle of
Attack (deg)

Velocity
(m/s)

Roll Angle
(deg)

Heading
angle (deg)

Flight Path
angle (deg)

6.00

-53.83

1.86

221.02

-55.70

A pulsed roll-manoeuvre is carried out on the descent trajectory to display the effectiveness of the
controllers. A pulse pattern is selected as the 3 commanded pulses are carried out in different atmospheric
conditions. This tests the performance of the scheduled-baseline controller where the desired dynamics are a
function of dynamic pressure.

30
American Institute of Aeronautics and Astronautics

Roll Angle Tracking

30

Commanded

Baseline
Augmentation

(deg)

20
10
0
-10

10

20

30
Time (s)

50

60

Altitude Profile

7000

Trajectory - Baseline
Trajectory - Augmentation

6000
Altitude (m)

40

5000
4000
3000
2000
1000

10

20

30
Time (s)

40

60

Control Surface Deflection - aileron

- Baseline
a

(deg)

50

- Augmentation
a

0
-2
-4
-6

10

20

30
Time (s)

40

50

60

Fig. 5. Roll Angle Tracking - Benchmark Case (L1 and L7)


In the benchmark case, the PPC is able to track the desired roll angle. In the presence of no uncertainties,

the

L1 augmented controller isnt able to improve the performance of the baseline controller significantly. This

means that the baseline controller has been tuned to stabilise the system and perform well under nominal conditions.

31
American Institute of Aeronautics and Astronautics

Roll Angle Tracking

40

Commanded

Baseline
Augmentation

30

(deg)

20
10
0
-10

10

20

30
Time (s)

50

60

Altitude Profile

7000

Trajectory - Baseline
Trajectory - Augmentation

6000
Altitude (m)

40

5000
4000
3000
2000
1000

10

20

30
Time (s)

40

60

Control Surface Deflection - aileron

- Baseline
a

(deg)

50

- Augmentation
a

0
-1
-2
-3

10

20

30
Time (s)

40

50

60

Fig. 6: Reduced aileron deflection (L3 and L9)


In the case of reduced aileron function, the baseline controller in Error: Reference source not found shows
significantly less oscillations compared to the

L1

augmented controller when only the PPC controller is

operating. Once the augmentation is turned on, although the steady state tracking ability of the controller improves,
the oscillations increase in the transient period.

32
American Institute of Aeronautics and Astronautics

Roll Angle Tracking

40

Commanded

Baseline
Augmentation

30

(deg)

20
10
0
-10

10

20

30
Time (s)

50

60

Altitude Profile

7000

Trajectory - Baseline
Trajectory - Augmentation

6000
Altitude (m)

40

5000
4000
3000
2000
1000

10

20

30
Time (s)

40

60

Control Surface Deflection - aileron

- Baseline
a

(deg)

50

- Augmentation
a

0
-2
-4
-6

10

20

30
Time (s)

40

50

60

Fig. 7. Asymmetric elevon deflection (L3 and L9)


The asymmetric deflection is defined as 10%. This means that there is a 10% difference in the deflection of
the left and the right elevon. The baseline controller for the first two commanded pulses shows inability to track the
desired value in the form of a steady state error. However, the

L1 augmented controller from the very start of the

simulation is able to track the desired roll angle. The importance of this test case lies in the realistic nature of the
uncertainty. Uncertainties are not likely to occur in a symmetric manner. Therefore, the improved performance with
the augmented controller is especially encouraging.

33
American Institute of Aeronautics and Astronautics

Roll Angle Tracking

30

Commanded

Baseline
Augmentation

(deg)

20
10
0
-10

10

20

30
Time (s)

50

60

Altitude Profile

7000

Trajectory - Baseline
Trajectory - Augmentation

6000
Altitude (m)

40

5000
4000
3000
2000
1000

10

20

30
Time (s)

40

60

Control Surface Deflection - aileron

- Baseline
a

(deg)

50

- Augmentation
a

0
-2
-4
-6

10

20

30
Time (s)

40

50

60

Fig. 8. Rudder failure: stuck at 1.0 degrees (L2 and L8)


Error: Reference source not found shows that the tracking performance of the baseline controller is
significantly worse than the augmentation in the case of a rudder failure. At no stage is the baseline controller able to
follow the commanded roll angle. The augmented controller, on the other hand, is able to track the desired roll angle,
with minimal oscillations in the transition period. As rudder failure is a major failure, this result is especially
encouraging as it is clear that the baseline controller is unable to cancel the rudder uncertainty in the system,
whereas the

L1 augmented controller is successful.

34
American Institute of Aeronautics and Astronautics

Roll Angle Tracking

40

Commanded

Baseline
Augmentation

30

(deg)

20
10
0
-10

10

20

30
Time(s)

50

60

Altitude Profile

7000

Trajectory - Baseline
Trajectory - Augmentation

6000
altitude (m)

40

5000
4000
3000
2000
1000

10

20

30
Time(s)

40

60

Control Surface Deflection

- Baseline
a

(deg)
a

50

- Augmentation
a

0
-2
-4
-6

10

20

30
Time(s)

40

50

60

Fig. 9. Gravimetric Errors (L4 and L10)


Error: Reference source not found highlights that the gravimetric uncertainties do not have a profound
impact on the performance of the PPC and the

L1

augmented controller. However, the

controller does improve the steady state tracking ability and the settling time of the PPC controller.

35
American Institute of Aeronautics and Astronautics

L1

augmented

Roll Angle Tracking

40

Commanded

Baseline
Augmentation

30

(deg)

20
10
0
-10

10

20

30
Time(s)

50

60

Altitude Profile

7000

Trajectory - Baseline
Trajectory - Augmentation

6000
altitude (m)

40

5000
4000
3000
2000
1000

10

20

30
Time(s)

40

60

Control Surface Deflection

- Baseline
a

(deg)
a

50

- Augmentation
a

0
-2
-4
-6

10

20

30
Time(s)

40

50

60

Fig. 10. Aerodynamic uncertainties (L5 and L11)


In the presence of aerodynamic uncertainties, once again the differences between the PPC and the

L1

augmented controller arent too pronounced. However, it does seem that the augmentation performs better in the
transition region and also has a reduced steady state error as compared to the baseline.

36
American Institute of Aeronautics and Astronautics

Roll Angle Tracking

40

Commanded

Baseline
Augmentation

30

(deg)

20
10
0
-10

10

20

30
Time(s)

50

60

Altitude Profile

7000

Trajectory - Baseline
Trajectory - Augmentation

6000
altitude (m)

40

5000
4000
3000
2000
1000

10

20

30
Time(s)

40

60

Control Surface Deflection

- Baseline
a

(deg)
a

50

- Augmentation
a

0
-2
-4
-6

10

20

30
Time(s)

40

50

Fig. 11. Aerodynamic uncertainties (L5 and L11)

37
American Institute of Aeronautics and Astronautics

60

Roll Angle Tracking

40

Commanded

Baseline
Augmentation

30

(deg)

20
10
0
-10

10

20

30
Time(s)

50

60

Altitude Profile

7000

Trajectory - Baseline
Trajectory - Augmentation

6000
altitude (m)

40

5000
4000
3000
2000
1000

10

20

30
Time(s)

40

60

Control Surface Deflection

- Baseline
a

(deg)
a

50

- Augmentation
a

0
-2
-4
-6

10

20

30
Time(s)

40

50

60

Fig. 12. Rudder failure: -2 to 2 degrees (Monte Carlo) (L2 and L8)
Error: Reference source not found depicts the performance of the PPC (blue) and the

L1 augmented

controller (red) when the rudder experiences a failure and it is stuck at an angle between -2 and 2 degrees. It is
evident that the baseline controller is unable to cancel out the effect of the rudder failure. Although the controller is
stable there is a large steady state error. The

L1 augmented controller is able to cancel out this uncertainty and is

able to successfully track the desired roll angle. This is particularly impressive when considering that for one of the
cases the baseline controllers steady state value is

350 (for the case when the commanded roll angle is 250

).

38
American Institute of Aeronautics and Astronautics

Roll Angle Tracking

60

Commanded

Baseline
Augmentation

(deg)

40
20
0
-20

10

20

30
Time(s)

50

60

Altitude Profile

7000

Trajectory - Baseline
Trajectory - Augmentation

6000
altitude (m)

40

5000
4000
3000
2000
1000

10

20

30
Time(s)

40

60

Control Surface Deflection

- Baseline
a

(deg)
a

50

- Augmentation
a

0
-2
-4
-6

10

20

30
Time(s)

40

50

60

Fig. 13. Rudder failure: -5 to 5 degrees (Monte Carlo) (L2 and L8)
For the case where the failed rudder is permanently stuck between -5 and 5 degrees, the baseline controller,
although stable, is unable to track the desired roll angle by a considerable amount. For one of the cases during the
first pulse, the actual roll angle (baseline) is close to

50

. Once again the

L1 augmented controller is able to

perfectly track the desired roll angle. It is seen that the control deflection is more when the augmentation is turned
on, as it is trying to correct the error.

39
American Institute of Aeronautics and Astronautics

Roll Angle Tracking

80

Commanded

Baseline
Augmentation

60

(deg)

40
20
0
-20

10

20

30
Time(s)

50

60

Altitude Profile

7000

Trajectory - Baseline
Trajectory - Augmentation

6000
altitude (m)

40

5000
4000
3000
2000
1000

10

20

30
Time(s)

40

60

Control Surface Deflection

- Baseline
a

(deg)
a

50

- Augmentation
a

0
-2
-4
-6

10

20

30
Time(s)

40

50

60

Fig. 14. Rudder Failure: -7.5 to 7.5 degrees (L2 and L8)
The most severe case of rudder failure presented in this thesis is when the rudder is stuck between -7.5 and
7.5 degrees. The PPC fails (although stable) to track the desired value. Despite the severe failure the

L1

augmented controller is able to track the desired roll angle. The aileron deflection increases for the augmented
controller in order to correct for this error.

40
American Institute of Aeronautics and Astronautics

Roll Angle Tracking

60

Commanded

40

Baseline
Augmentation

(deg)

20
0
-20
-40
-60

10

20

30
Time(s)

50

60

Altitude Profile

8000

Trajectory - Baseline
Trajectory - Augmentation

6000
altitude (m)

40

4000
2000
0

10

20

30
Time(s)

40

60

Control Surface Deflection

- Baseline
a

(deg)
a

50

- Augmentation
a

0
-2
-4
-6

10

20

30
Time(s)

40

50

60

Fig. 15. Rudder Failure: -7.5 to 7.5 degrees (L2 and L8)
The most realistic case in terms of dealing with uncertainties is placed in Error: Reference source not
found. After re-entry the glider will most likely have a multitude of uncertainties. None of the uncertainties
mentioned up to this point occur as an isolated event. The uncertainties that are introduced into the system include
aerodynamic uncertainties, gravimetric uncertainties, actuator uncertainties (asymmetric elevon deflection) and
rudder failure (-5 to 5 degrees). As depicted in Error: Reference source not found, the PPC, although stable, is
unable to track the desired roll angle. Under some conditions, the value of the roll angle reaches as high as
and as low as

500

55 0 . The L1 augmented controller, however, is able to track the commanded roll angle even

in the presence of such uncertainties and rudder failure.

41
American Institute of Aeronautics and Astronautics

Main Conclusions from Roll Angle Tracking


The main conclusions that are drawn from the test cases presented in Error: Reference source not found (lateral
test cases) are as follows:

The

L1 augmented controller is able to marginally improve the performance of the baseline controller

in the presence of gravimetric and aerodynamic uncertainties.

The

L1 augmented controller is able to improve the tracking ability and in turn reduce the steady state

error in the presence of asymmetric elevon deflection.

The

L1 augmented controller significantly improves the performance of the baseline controller when

there is major rudder failure (Error: Reference source not found Error: Reference source not found). The
PPC is unable to follow the commanded roll angle, whereas the augmented controller displays minimal
steady state error when there is a rudder failure.

There is a significant improvement in the performance and tracking ability of the controller when the
augmentation is turned on in the presence of multiple uncertainties. As seen in Fig 5.38 the augmentation is
able to track the commanded roll angle even in the presence of significant uncertainties.

E. Performance Analysis of Control Methodologies


This sub-section of the paper presents the

L2norm of the tracking error,

e pL

. This norm is used as

a way to compare the performance of the baseline to the augmented controller. The tracking error norm is defined
as:

TS

e pL = |e p| dt
2

The table below outlines the

e pL

(124)

for all the simulation cases presented in Table 2. The performance

results from the Monte Carlo runs are placed in Table 5. The percentage in the tracking error norm is also placed in
Table 5 below.
Table 5. Roll Tracking Error Norm
Baseline

Augmentation

Percentage
Improvement (%)

Baseline

26.42

25.51

3.42

Rudder failure (1.0 deg)

37.89

25.22

33.43

Rudder failure (-2 to 2 deg)

32.70

25.21

22.91

42
American Institute of Aeronautics and Astronautics

Baseline

Augmentation

Percentage
Improvement (%)

Rudder Failure (-5 to 5 deg)

52.74

25.27

52.08

Rudder Failure (-7.5 to 7.5 deg)

71.73

25.37

64.64

Reduced elevon function

36.25

31.63

12.74

Asymmetric elevon deflection

30.87

25.93

16.02

Gravimetric uncertainties

26.21

25.37

3.24

Aerodynamic uncertainties

26.53

25.65

3.34

Combination of errors

112.20

26.99

75.94

For the all the test cases, there is an increase in the aileron deflection (therefore in turn the elevon deflection)
when compensating for the errors that are present in the system. This leads to an increase in the control surface
metric in the lateral/direction case as is seen in Table 5. 2.

e pL

The

for the side slip angle showed minimal change therefore it hasnt been presented here. The

commanded sideslip angle is

00 . There are oscillations in the sideslip angle only when the roll controller is

carrying out the roll manoeuvre. The tracking error norm for the roll angle is placed in Error: Reference source not
found. The main conclusions are as follows:

For the benchmark case, the

L1

augmented controller is able to reduce the tracking error norm;

however the change is not significant.

The most pronounced changes in the tracking error norm are the rudder failure cases. When there is a
rudder failure, the norm is significantly worse when only the PPC is operating. The augmentation is
able to significantly reduce this tracking error to values close to the benchmark case.

In the case of reduced elevon function, where the deflection is limited to only

60 , although the

tracking error norm is reduced, it is not reduced to the value that is the case in the benchmark scenario.

The most important improvement is when there is a combination of errors. The tracking error norm

L1

when only the baseline is operating is 112.20. However, when the

augmented controller is

operating, the tracking error norm is 26.99. This result is particularly encouraging and shows that in the
presence of vast amounts of uncertainties and system failures, the

L1 augmented controller gives

extremely promising improvements in terms of the tracking performance.

43
American Institute of Aeronautics and Astronautics

The same control surface metric is utilised to analyse the deflection of the elevons and the rudder for the
lateral/directional dynamics.
Table 5. 1 presents the mean and the standard deviation of the altitude for the failure cases for the
lateral/directional dynamics. For the case where there are aerodynamic and gravimetric errors, there is little
improvement in terms of the standard deviation of the altitude when the

L1 augmented controller is running.

However, in the presence of rudder failure and also a combination of errors the

L1

augmented controller

significantly improves the standard deviation significantly. The mean altitudes for both the configurations do not
vary considerably.
Table 5. 1. Altitude Mean and Standard Deviation (Lateral Test Cases)
Baseline Mean (Standard
Deviation) meters

Augmentation Mean (Standard


Deviation) - meters

Rudder failure (-2 to 2 deg)

1433.28 (10.65)

1437.35 (1.76)

Rudder Failure (-5 to 5 deg)

1427.18 (31.20)

1437.25 (1.93)

Rudder Failure (-7.5 to 7.5 deg)

1418.35 (53.81)

1437.15 (2.92)

Gravimetric uncertainties

1463.94 (39.63)

1466.56 (39.79)

Aerodynamic uncertainties

1076.92 (2.29)

1079.72 (2.28)

Combination of errors

933.70 (42.06)

957.69 (14.88)

Table 5. 2. Control Surface Metric

Baseline (

Mu
b

Augmentation

(Mu

L1

Percentage
Change (%)

Rudder failure (1.0 deg)

0.75

1.19

58.67

Rudder failure (-2 to 2 deg)

0.49

1.08

124.41

Rudder Failure (-5 to 5 deg)

1.15

1.55

34.78

Rudder Failure (-7.5 to 7.5 deg)

1.70

2.03

19.41

Reduced elevon function

0.77

0.89

15.58

Asymmetric elevon deflection

0.21

0.68

223.81

Gravimetric uncertainties

0.11

0.81

636.36

44
American Institute of Aeronautics and Astronautics

Aerodynamic uncertainties

0.12

0.67

458.33

Combination of errors

2.01

2.15

6.97

The main conclusions drawn from Table 5. 2 are as follows:

All of the values for

Mu

L1

were greater than

Mu

. As the system was decoupled, the ailerons, and

therefore in turn the elevons, has to be deflected more to account for the uncertainties in the roll channel.

As expected, the highest values for

Mu

L1

are for the rudder failure cases, where the steady state errors

were the highest.

The highest value for

Mu

L1

is when there is a combination of errors present in the system.

The sideslip angle tracking of the controllers is depicted in Fig 5. 1.

45
American Institute of Aeronautics and Astronautics

Sideslip Angle Tracking

Commanded

Baseline
Augmentation

(deg)

1
0
-1
-2

10

20

30
Time (s)

50

60

Altitude Profile

7000

Trajectory - Baseline
Trajectory - Augmentation

6000
Altitude (m)

40

5000
4000
3000
2000
1000

10

20

30
Time (s)

40

50

60

Control Surface Deflection - aileron

- Baseline
r

(deg)

- Augmentation
r
0

-5

10

20

30
Time (s)

40

50

60

Fig 5. 1. Sideslip Angle Tracking - Benchmark Test Case


The only time the sideslip angle change is when there is a commanded roll angle. Once the roll angle stabilizes
and reaches steady state, the sideslip angle settles at zero.

46
American Institute of Aeronautics and Astronautics

Sideslip Angle Tracking

Commanded

Baseline
Augmentation

(deg)

2
1
0
-1
-2
-3
0

10

20

30
Time(s)

40

50

60

Altitude Profile

7000

Trajectory - Baseline
Trajectory - Augmentation

6000

altitude (m)

5000
4000
3000
2000
1000
0

10

20

30
Time(s)

40

50

60

Control Surface Deflection

- Baseline
r

- Augmentation
r

(deg)
r

2
0
-2
-4
-6

10

20

30
Time(s)

40

50

60

Fig 5. 2. Combination of Errors Sideslip Angle


The main limitation of SISO is that the functioning ailerons does not have an effect on the sideslip angle when
there is a rudder failure. This aspect of the behaviour of the controller synthesis is highlighted in Fig 5. 2. Even with
the augmentation on, the sideslip angle is not tracked in the case of a rudder failure. Therefore, for future work the
application of a MIMO baseline controller to control the lateral/directional dynamics is to be considered. With a
MIMO controller, the oscillations and steady state errors in the sideslip angle channel could be better cancelled out
when a roll angle is commanded. Furthermore, when there is a rudder failure, the functioning aileron cant cancel out
the effects of the failed rudder has on the sideslip angle. This is due to the limitations introduced at the modelling
stage as a result of the decoupling.
47
American Institute of Aeronautics and Astronautics

V.

Conclusion

This paper presents a comparative study of the performance and robustness of two types of controllers; the
SAPPC and the SAPPC augmented with the

L1

adaptive controller. The controller structure of both

methodologies is presented. The pole placement controller with perfect state feedback and in the presence of no
uncertainties is able to carry out the desired manoeuvre. The roll controller and the side slip controller performed
well however the performance of the baseline controllers significantly reduced in the presence of uncertainties. The

e pL

presented in this paper shows the improvement that the augmentation provides in terms of performance

and also robustness. For all the uncertainty cases the augmentation configuration reduced the

e pL

as

compared to the baseline controller. Full control authority is maintained for the test trajectory as there is no control
surface saturation. Future work will concentrate on the implementation of the control law that compensates the
unmatched uncertainties. This should lead to a further reduction in the tracking norm when the

L1

augmentation is used. For further studies the atmosphere model will also be turned on to test the performance and
robustness of the controllers to such disturbances.

References
[1]. Zipfel, P., Modeling and Simulation of Aerospace Vehicle Dynamics, 2nd ed., AIAA-Education Series, Virginia, 2007.
[2]. Hovakimyan,N. and Cao,C.,

L1

Adaptive Control theory: Guaranteed Robustness with Fast Adaption, SIAM,

Philadelphia, PA, USA, 2010.


[3]. Somanath, A., and Annaswamy, A., Adaptive Control of Hypersonic Vehicle in the Presence of Aerodynamic and Centre of
Gravity Uncertainties, 49th IEEE Conference on Decision and Control, Atlanta, 2010, pp. 4661-4666.
[4]. Xu, H., Ioannou, P., and Mirmirani, M., Adaptive Sliding Model Control Design for a Hypersonic Flight Vehicle, Journal
of Guidance Control and Dynamics, Vol, 27, No. 5, 2004, pp. 829-838.
[5]. Bin, X., DaoXiang, G., and Wang, S., Adaptive Neural Control Based on HGO for Hypersonic Flight Vehicles, Science
China Information Sciences, Vol. 54, No. 3, 2011, pp. 511-520.
[6]. Yuang, Z., and Feng, Y., Disturbances Analysis for a Hypersonic Vehicle Based on Adaptive Fuzzy Control, Eighth
International Conference on Fuzzy Systems and Knowledge Discovery, Vol. 1, Shanghai, 2011, pp. 123-128.
[7]. Creagh, M., Beasley, P., and Kearney,M., Adaptive Control for a Hypersonic Glider using Parametric Feedback from System
Identification, AIAA Guidance, Navigation and Control Conference, Portland, 2011.
[8]. Bolender, M., An Overview on Dynamics and Controls Modeling of Hypersonic Vehicles, American Control Conference,
St Louis, 2009, pp. 2507-2512.
[9]. Coleman, C., and Faruqi, F., On Stability and Control of Hypersonic Vehicles, Weapons Systems Division Defence
Science and Technology Organisation, URL: http://www.dtic.mil/cgi-bin/GetTRDoc?AD=ADA521248 [cited 7 January
2013].
[10].Baumann, E., Bahm, C., Strovers, B., Beck, R., and Richard, M., The X-43A Six Degree of Freedom Monte Carlo
Analysis, NASA-H-2683, 2007.
[11]. Lei, Y., Cao, C., Cliffe, E., Hovakimyam, N., Kurdila, A., and Wise, K.,

L1

Adaptive Controller for Air-Breathing

Hypersonic Vehicle with Flexible Body Dynamics, American Control Conference, Conference Publication, St. Louis, 2009,
pp. 3166-3171.
[12].Syvertson, C., Gloria, H., and M. Sarabia., Aerodynamic Performance and Static Stability and Control of Flat-top
Hypersonic
Gliders
at
Mach
Numbers
from
0.6
to
18.
URL:
https://ia600505.us.archive.org/5/items/nasa_techdoc_19660024021/19660024021.pdf [cited 9 November 2013].
[13].Rainey, R., Static Stability and Control of Hypersonic Gliders. UNT Digital Library. URL:
http://digital.library.unt.edu/ark:/67531/metadc52933/ [cited 10 November 2013].
[14].Griffin, B., Burken, J., and E. Xargay.,
Lateral/Directional

Dynamics:

L1

Adaptive Control Augmentation System with Application to X-29


Multi

Input

Multi-Output

Approach.

http://ntrs.nasa.gov/archive/nasa/casi.ntrs.nasa.gov/20100037212_2010040718.pdf [cited 28 October 2013].

48
American Institute of Aeronautics and Astronautics

URL:

[15].Banerjee,S., Creagh, M., Boyce, R., Baur, B., Zhang, Z., and F. Holzapfel.,

L1

Adaptive Control Augmentation

Configuration for a Hypersonic Glider in the Presence of Uncertainties. AIAA SciTech Guidance, Navigation and Control
Conference, 2014, National Harbour, Maryland.
[16].Ito, D., Georgie, J., Valasek, J., and D. Ward., Reentry Vehicle Flight Control Design Guidelines: Dynamic Inversion.
URL: http://ston.jsc.nasa.gov/collections/trs/_techrep/TP-2002-210771.pdf [cited 11 March 2014].
[17].Young, A., Cao, C., Hovakimyan, N., and E Lavretsky., An Adaptive approach to nonaffine control design for aircraft
applications, AIAA Guidance, Navigation and Control Conference and Exhibit, Keystone, CO, USA, August 2006.
[18].Fichter, W and W Grimm., Flugmechanik, Skript: Institute for Flight Mechanics and Flight Control, University of Stuttgtart.

49
American Institute of Aeronautics and Astronautics

Anda mungkin juga menyukai