Aeroelastic flutter non linear control

Professor Mario Di Bernardo

Faculty of engineering, Naples
Masters degree in control engineering

Giovanni Pugliese Carratelli M58/30

"Heavier than air flying machines are impossible"

Lord Kelvin

Summary
The purpose of this document is to investigate some non linear control strategies for a 2 degree of freedom
(DOF) wing section subject to aero elastic utter. In the beginning of the document it will be shown what aero
elastic utter is with some examples. A mathematical model is then shown, the BACT model, with system some
analysis. After the system analysis results, two control strategies are developed and results and simulations are
shown.

Contents
Summary

1 Model and system analysis

1.1 What is aereoelastic uttering . . . . . . . . . . . . . . . .
1.1.1 Examples . . . . . . . . . . . . . . . . . . . . . . .
1.2 The NASA BACT model . . . . . . . . . . . . . . . . . .
1.2.1 BACT MODEL with two control surfaces . . . . .
1.3 State space representation and system open loop analysis
1.3.1 State space representation . . . . . . . . . . . . . .
1.3.2 Open loop analysis . . . . . . . . . . . . . . . . . .

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

9
9
11
14
16
16
16
19

2 Input Output Feedback linearization

2.1 Pitch FBL control . . . . . . . . . . . . . .
2.1.1 Simulations . . . . . . . . . . . . . .
2.1.2 Hidden dynamics analysis . . . . . .
2.1.2.1 Stability analysis . . . . . .
2.2 Plunge FBL control . . . . . . . . . . . . .
2.2.1 Simulations . . . . . . . . . . . . . .
2.2.2 Hidden dynamics analysis . . . . . .
2.2.2.1 Lyapunov stability analysis
2.2.2.2 Bifurcation analysis . . . .

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

25
25
27
33
33
34
35
36
37
38

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

43
43
44
44
45
45
45
46
46
48
49
53
57
60
61
62
63
67

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.

3 MRAC Minimum Controller Synthesis control for flutter

3.1 MCS algorithm . . . . . . . . . . . . . . . . . . . . . . . . .
3.1.1 MCS extensions . . . . . . . . . . . . . . . . . . . .
3.1.1.1 MCS-LQ . . . . . . . . . . . . . . . . . . .
3.1.1.2 MCSI . . . . . . . . . . . . . . . . . . . . .
3.1.1.3 EMCS . . . . . . . . . . . . . . . . . . . . .
3.1.1.4 NEMCS . . . . . . . . . . . . . . . . . . . .
3.1.1.5 LQ-NEMCSI . . . . . . . . . . . . . . . . .
3.2 MCS control synthesis and simulations . . . . . . . . . . . .
3.2.0.6 Simulations . . . . . . . . . . . . . . . . . .
3.2.0.7 MCS controller . . . . . . . . . . . . . . . .
3.2.0.8 MCSI controller . . . . . . . . . . . . . . .
3.2.0.9 EMCSI controller . . . . . . . . . . . . . .
3.2.0.10 NEMCSI controller . . . . . . . . . . . . .
3.2.0.11 LQ-NEMCSI controller . . . . . . . . . . .
3.2.0.12 Gain-locking . . . . . . . . . . . . . . . . .
3.2.0.13 Velocity variation rejection and comparison
3.2.0.14 Hybrid parameter variation . . . . . . . . .
7

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

3.2.0.15 Chaos recovery . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

69

Bibliography

72

List of figures

73

List of tables

79

Chapter 1

Model and system analysis

It is possible to fly without motors, but not without knowledge and skill. This I conceive to be fortunate, for man, by
reason of his greater intellect, can more reasonably hope to equal birds in knowledge than to equal nature in perfection
of her machinery.
Wilbur Wright, Letter to the Western society of engineers, 1900

1.1

What is aereoelastic fluttering

Aeroelasticity is the interaction of structural, inertial and aerodynamic forces. It occurs to systems subject to an
airstream (or more generally in a uid stream), for example to airfoils or even bigger structures such as bridges
or buildings. Aeroelasticity is under certain conditions characterised by what is called flutter. Aeroelastic utter
is an oscillatory aeroelastic instability characterized by the loss of elastic recall and low damping due to the
presence of aerodynamic loads. In the aerospace industry this is a very well known problem as it happens to
occur to airplanes wings.
The conditions under which utter can be observed are various and depend mainly on: the speed at which the
structure is moving in the uid, the elastic recall to which the foil is subject to (as the foil is a structure it has
an elastic behaviour) and the angle between the uid and the foil (also known as angle of attack AoA). Indeed,
in the case of an airplane, wing structural deformation leads to higher aerodynamic forces making utter a
self-feeding mechanism that may lead to catastrophe, moving so from an equilibrium point to a Limit Cycle
9

Model and system analysis

Oscillation (LCO) or to chaos. If the damping is not adequate, the imbalance between energy input and the
structural dissipation will result in very large oscillations or unconstrained increase of amplitude.
A simple example can show qualitatively what happens: consider rst a foil in a steady airstream. Such foil in
a uid stream is subject to lift, air resistance and a pitch moment as can be seen in Fig.1.1 .

Figure 1.1: Lift (L), resistance (R) and pitch moment (P) of a foil in a steady airstream

Lift, resistance and pitch moment depend, among other factors, on the square of the airstream velocity, the
exposed surface to the airstream and the AoA, .

Functions Cl ,Cr and Cm

are depicted in Fig.1.2.

1
L = Cl () U 2 Cx A
2
1
R = Cr () U 2 Cx A
2
1
P = Cm () U 2 Cx A
2
are non linear functions of and are dierent for every foil. Some typical behaviours

Figure 1.2: Typical Cl ,Cm and Cr behaviour

A simple model to show how LCO occur on such system is to consider a 2nd order torsional ODE as follows:
Consider now a foil, that is immerged in an airstream and is subject also to an elastic recall K, for example a
wing section like as Fig.1.3.

10

1.1. What is aereoelastic fluttering

Figure 1.3: Foil subject to elastic recall K

A simple model to show how LCO occur on such system is to consider a 2nd order torsional ODE as follows:
I
+ c + K () = M (, )

(1.1)

M (, )
is the pitch moment, c ()
is the damping term and K () is the torsional elastic recall. Assuming
that M is only function of , possible for low velocities, the model can be recast to the following form:
I
+ c + (K () M ()) = 0

(1.2)

When hit by an airstream at a critical speed U0 the foil will be subject to a pitch moment increase, that will
lead to higher values of the AoA. Pitch angle will increase if K () M () < 0; so the structure elastic recall
force plays a crucial role. Indeed when for a certain value 0 of AoA, after which K () M () > 0, the elastic
recall will be greater than pitching moment and bring the foil towards = 0. As the system damping is low,
which is typical for structures, overshooting from = 0 will occur. This will result in a negative AoA and an
increasing pitch moment in opposite direction as shown in Fig.1.4

Figure 1.4: Foil subject to elastic recall K in opposite direction after overshoot

After overshooting, very much as what happens for positive AoA, the structures recall is lower than pitching
moment and so there will be a negative increase of AoA. Since the airstream is steady, pitching moment will grow
until it is greater than the elastic recall for negative values of , increasing the AoA until K () M () > 0.
The major elastic recall over the pith moment will bring the system towards = 0, and an overshoot will occur
again for positive values of the AoA. A a self feeding mechanism will so begin, this is called flutter.
This example shows how utter is in this simple dynamical system a LCO. More complex models are of course
possible with more than one degree of freedom where chaotic behaviour is possible.

1.1.1

Examples

Flutter is not only limited to the aerospace industry, it can indeed occur to other structures in a uid, as for
instance chimneys and bridges, or even simply sign posts. Although the focus of this report is towards airplane
(or gliders) wing active uttering control, in the past many other ways to avoid the phenomena have been
successfully applied in aeronautics (i.e. passive or structural uttering control). The main strategy along this
line is to dimension the structure in such a way that the energy introduced in the system is well damped in
11

Model and system analysis

normal use (mass distribution is the main parameter on which to work).
Some examples of what utter can lead to are reported in the following. As for the aerospace industry an
example of wing utter was shown in a test ight by NASA in 1966. The test ight pilots brings the Piper
PA-38 to speed that causes utter and slows down just before any structural failure.

Figure 1.5: The tail of the Piper during an LCO

Some accidents have happened due to uttering, the rst example is the Tackoma bridge in 1940. The bridge
was so that it would oscillate at its natural mode when subject to wind at approximately 67mph. When this
occurred the bridge structure failed as shown by these impressive images.

In recent times some studies have qualitatively related how utter and dry Coulomb friction are a close
phenomena as shown in [2]. In this paper a simple 2 Degree of Freedom (DoF) mechanical arm is built with
two elastic hinges on which a load is applicable. The arm is shown in Fig., and known as Ziegler Column.

12

1.1. What is aereoelastic fluttering

Figure 1.6: The Ziegler Column 2 DoF arm

This is a two-degree-of-freedom structure made up of two rigid bars, internally jointed with an elastic
rotational spring, externally xed at one end through another elastic rotational spring, while at the other end
subject to a tangential follower load P coaxial to the rod to which it is applied as schematically show in 1.7.
One side of the arm is xed to a and on the other a wheel is mounted so to create a follower force P with a
movable metal plate that allows friction between the wheel and the plate.

Figure 1.7: Schematic representation of the 2 DoF mechanical arm. vp is the plate velocity

It is shown that the structure becomes dynamically unstable at a certain load level , so that it evidences
utter and, at higher load, divergence instability. In Fig.1.8 it is quite clear how the system moves towards a
LCO both from the model predictions and the experimental data.

Figure 1.8: Experimental data and model prediction for the Ziegler Column

13

Model and system analysis

1.2

The NASA BACT model

In the past many approaches have been used to model and actively control utter. The major modeling results
are by Theodorsen who developed the classical unsteady aerodynamic theory which accounts for the aerodynamic
damping at dierent conditions, and showed how utter depends on it. After Theodorsen, Mukhopadhyay and
Gangsaas created 20th and 50th order models,and strategies for order reduction. They used state feedback as
the control method, and implemented estimators to describe unmeasured states. They employed proportional
gain feedback methods developed from root locus plots. A quasi-steady aerodynamic model coupled with a two
degree-of-freedom structural model was used to develop several types of feedback. The development directly
feeds one of four variables to the control surface through a proportional gain.
Aeroelastic systems typically contain nonlinearities which are either neglected or simplied to a linear form
for analysis. Nonlinearities which occur in aeroelastic systems include control saturation, free play, hysteresis,
piece-wise linear, and continuous nonlinearities.
Later on it was shown that a poor agreement between theory and experiment in utter is most likely due to
nonlinear structural elastic models. So detailed examination of many types of nonlinearities that may aect
aeroelastic systems is presented in various articles. Tang and Dowell introduced a free play nonlinearity in the
torsional stiness and examined the nonlinear aeroelastic response. For various initial condition they show that
LCO is dependent upon free stream velocity, initial pitch condition, magnitude of the free play nonlinearity and
initial conditions.
One the major developments was given by NASA in 1997, by building the Benchmark Active Control Technology
(BACT). It is a two degree of freedom model where the pitching movement and the plunging one, are respectively
restrained by a pair of springs attached to the elastic axis(EA)of the airfoil. A prototype was also built as shown
in Fig.1.9 where one or two trailing-edge control surfaces are used to control the air ow, thereby providing
maneuverability to suppress instability. The BACT model is accurate for airfoils at low velocity and has been
conrmed by wind tunnel experiments.

Figure 1.9: Configuration of the nonlinear 2-D prototypical aeroelastic wing

In this report the BACT model will be shown, thus with respect to the example in equation 1.2, a second
degree of freedom known as plunge is taken into account. Plunge is introduced so to consider also apping of the
considered wing section. The model is a simple representation of an aeroelastic system for low speed, where all
non linear terms from experimental data are taken into account as shown in [9]. Hence the equations governing
the motion of the aereoelastic system are:
"

m
mx b

#(
) "

mx b
h
ch
+
I

0
c

#(

"
Kh
+
0

0
K ()

#(

L
M

(1.3)

where h is the plunge motion and is the pitch or AoA. In equation 1.3, m is the mass of the considered section
of the wing, and I is the mass moment of inertia about the elastic axis. The position of the elastic axis with
respect to the center of mass of the considered wing section can be varied and is referred as x . Constants ch
and c are linear damping constants of the system. Kh is the spring constant for the plunge motion and K ()
is the non linear stiness of the pitch motion. In this report the non linear stiness K () is assumed to be a
14

1.2. The NASA BACT model

4th order polynomial function:
K () =

4
X

Ki i = K0 + K1 + K2 2 + K3 3 + K4 4

(1.4)

i=0

Figure 1.10: Wing cross-section representation

M and L, in case of a single control surface as in Fig. 1.10, are respectively the input moment and the
quasi-steady aerodynamic lift and are modeled as [3]:
L = U 2 bcl ( +

h
1

+ ( a)b ) + U 2 bcl
U
2
U

(1.5)

h
+ ( a)b ) + U 2 b2 cm
(1.6)
U
2
U
where is air density, U is the air stream velocity, is the angle between the foil and the trailing edge control
surface. cl and cm are the lift and momentum coecients for the AoA and cl and cm are respectively the
lift and moment coecients for the control surface. a is the distance between mid-chord1 and the elastic axis
(EA) as shown in Fig.1.11.
M = U 2 b2 cm ( +

Figure 1.11: Wing cross-section schematic representation showing a and mid-chord b

After substituting the quasi-stead forces from equations 1.5 and 1.6 into equation 1.3:
"
#(
#(
)
) "

m
mx b
h
U b2 cl ( 12 a)
h
ch + U bcl
+
+
mx b
I
c U b3 cm ( 12 a)
U b2 cm

#(
) (
)
"
h
bcl
Kh
U 2 bcl
=
U2

b2 cm
0 U 2 b2 cm + K ()

(1.7)

1 Chord is the imaginary line joining the trailing edge and the center of curvature of the leading edge of the cross-section of an
airfoil

15

Model and system analysis

It is important to note that the only source of non linearity is given by the stiness and an extension for higher
velocity of the model is possible by considering the pith moment as a quadric function of .

1.2.1

BACT MODEL with two control surfaces

A extension of the model presented in equation 1.7 can be derived for when two control surfaces are available
on the trailing edge as depicted in Fig. 1.12 The model can be derived considering the following lift and pitch:

Figure 1.12: Two trailing edge control surfaces

L = U 2 bcl ( +
M = U 2 b2 cm ( +

h
1

+ ( a)b ) + U 2 bcl2 2 + U 2 bcl2 2

U
2
U

(1.8)

h
+ ( a)b ) + U 2 b2 cm1 1 + U 2 b2 cm2 2
U
2
U

(1.9)

That substituted in equation 1.3 leads to:

"
#(
) "
#(
)

m
mx b
U b2 cl ( 12 a)
h
ch + U bcl
h
+
+
mx b
I

c U b3 cm ( 21 a)

U bcm
)
#(
#(
) "
"
1
h
bcl1 bcl2
Kh
U 2 bcl
U2
=
+
2
b2 cm1 b2 cm2

0 U 2 cm K ()

1.3
1.3.1

(1.10)

State space representation and system open loop analysis

State space representation

It is convenient for the analysis that follows, and for control design to have a state space formulation (SS) of
the system in equations 1.7 and 1.10. For this purpose the following state variable vector is used:

x1

x
2
x=
=

x3

x4

, x 4

(1.11)

and is the control input.

The SS formulation expresses the system in the following ane form:
x = f,x (x) + g(x)
where = U 2 and it is to note the equations are dependant on the airstream velocity and the elastic axis
location x . The notation with the two subscripts and x emphasizes the system dependance on the two
16

1.3. State space representation and system open loop analysis

parameters and the solutions are in fact a two-parameter family of solutions.
It is possible to obtain a SS representation of an equation in the following form:
Mx
+ Dx + Kx = F u(t)
by using the well know formula:
" # "
x 1
0
=
x 2
M 1 K

I
M 1 D

#" # "
#
x1
0
+
x2
M 1 F

In the considered case matrix M 1 will be:

M

"

I
d
mx b
d

mx b
d
m
d

so a synthetic SS formulation with a single trailing edge control surface is the following:

x3
0

0
x4
, g(x) =
f,x (x) =

g3

k1 x1 (k2 + p(x2 ))x2 c1 x3 c2 x4

k3 x1 (k4 + q(x2 ))x2 c3 x3 c4 x4

g4
In the following table the system term of equation 1.12 are explicated:

(1.12)

d = m(I mx2 b2 )
I (ch +Ubcl )
d
mx b(ch +Ubcl )+mUb2 cm
c3 =
d
b
k (x2 )
p(x2 ) = mx
d
k1 = Idkh
k3 = mxdbkh
I cl mx b3 cm
g3 =
d

c1 =

I Ub2 cl ( 21 a)mx bc +mx Ub4 cm ( 12 a)

d
mx Ub3 cl ( 21 a)+mcl mUb3 cm ( 21 a)
c4 =
d
q(x2 ) = m
d k (x2 )
I c +mx b3 cm
k2 = l d
mx b2 cl mb2 cm
k4 =
d
mx b2 cl +mx b2 cm
g4 =
d

c2 =

Table 1.1: System SS variables

The parameters for the following investigation are obtained from [8]. The non linear elastic recall, dened
in 1.4, that will be used is dened as follows:
K () = 2.82(1 22.1 + 1315.52 + 85803 + 17289.74)
The other parameters used are shown in the following table: Most of the calculations and integrations for the
m = 12.3870Kg
rcg = 0.0873 (b + ab)
b = 0.135m
N
kh = 2844.4 m
cl = 6.28
cm = ( 21 + a)cl

I = 0.065Kgm2
rcg
b
Kg
= 1.118 m
3
Ns
ch = 27.43 m
cl = 3.358
cm = 0.635

Table 1.2: System SS parameters

system have been carried out using the MATLAB/SIMULINK environment. At this purpose in Fig.1.13 the
SIMULINK block diagram is shown; it is possible to any shape to the parameters signals of air stream (U),
density of air () and Elastic Axis location (a).

17

Model and system analysis

[k2]
[x1]
[U]

Goto

[d]
From69

[x3]
Goto2

Product13

[x1]

From55[c1]
From4

Ka

[x2]

Goto3

From56 [c2]
[g3]
[U]

From58

Add

From57

Math
Function1
[x1]

Product5

[U]

Ka1

From10

[d]

[x2]

[x3]
From63[c3]
From61
[x4]
[c4]
From64

m
Add3
a7

u
Math
Function

Product9

From3
Product12

From1

[k3]

From2

From65

Math
Function3

From66

P(u)
O(P) = 4

[x2]

Product4

From5

[k4]

1
beta

x3_dot

Product2

[x4]

From9

Product1

Product10
Product11

x4_dot

Product7

From62

Product8
Add1

From11
[g4]
[U]

Product6

From59

Math
Function2

From60

[g3]
[a]

xa

From13
a

[a]

a1

[xa]
[d]

Goto4

Goto9

From48

xa

[rho]

rho

g3

From49
rho

[rho]

a2

a3

U
a4

Scope8
xa

[a]

a cma

From14

[xa]

[cma]

g3

From50

Goto6

cma

Scope10

[U]
[d]
a5

[xa]

xa

From12

[d]

From15

[k1]

k1

Goto5

Goto7
k1

[d]
From16

[rho]

[k2]

Goto13

rho

[xa]

[d]
From22

xa

From17

[xa]

cma

From18

Goto14

[rho]
rho

Scope7
[cma]

[k4]

d
k2

From19

From23

k4
xa

From24

k2

Scope2
[cma]

[g4]

cma

From25
[k3]
[d]

Goto10

k4

[d]

From51

Goto12
k3

From20

[rho]

xa

Scope

[xa]
From21

rho

g4

From52

k3

Scope9
xa

[xa]
g4

From53
[d]

[d]

[c1]

From26

[rho]

[xa]
From28

a6

From27

xa

[cma]

From37

[rho]

c1

From38
Scope3

cma
U

From31 [rho]

Goto8

[d]

xa
cma

Scope4

From36

[rho]

c3

Scope5

cma
U

c3

[c4]

rho

From43

[xa]

Goto11

xa

c4

From44
[a]

xa

From42
c2

From34

a8

[c2]

rho

From32

From33 [cma]

From35

[U]

rho

From41

[d]

[U]

From40
c1

From30

[xa]

[xa]
From39

[cma]
From29
[U]

[c3]

rho

[U]

Scope6
c2

From46

[cma]
From45

[a]
From47

a
cma

c4

Figure 1.13: Simulink block diagram for the systems open loop dynamics

18

x_dot

1
s
Integrator

[x3]

Product

Add2

Product14

From7

Product3

From54

P(u)
O(P) = 4

From8

[x4]
[k1]

From

a9
[xa]
u
b
Math
From70
Function5
a10

[x2]

[x4]

From6

Math
1 Function4

From68

[x2]
Goto1

[x3]

From67

1
x

1.3. State space representation and system open loop analysis

1.3.2

Open loop analysis

A rst simulation of an open loop response of the system is carried out considering the air ow velocity U = 15 m
s
and a = 0.4 with the following initial conditions = 0.1[rad], h = 0.1[m] and = h = 0 as depicted in Fig.
1.14a.
150

0.15

0.1

100

0.05

x4 [deg/s]

x4 [deg/s]

50

0.05

50
0.1

100

150
6

0.15

(x2) [deg]

0.2
0.015

0.01

0.005

(x2) [deg]

0.005

0.01

(a) AoA Phase portrait with = 0.1[rad], h = 0.1[m] and

(b) Plunge phase portrait with = 0.1[rad], h = 0.1[m]

,a = 0.4
= h = 0,U = 15 m
s

,a = 0.4
and = h = 0,U = 15 m
s

The system clearly shows a LCOs due to the structural non linearity in equation 1.4 and due to the low
damping. The suppression of the LCO and of a possible chaotic behaviour will be in the following chapters, one
of the main goals for the controller.
Other simulation are for dierent initial conditions. It is interesting to notice that the system other than LCO
the
behaviour also shows a chaotic behaviour 2 for some initial conditions. Indeed for low values of h and h,
system exhibits chaos as shown in the following gures.

0.6

0.4
0.4

0.2
0.2

x4

X3

0
0

0.2

0.2
0.4

0.4
0.6

0.1

0.05

0.05

0.1

0.15

(a) AoA chaotic Phase portrait with = 0.1[rad], h = 0[m]

a = 0.4
and = h = 0,U = 25 m
s

0.05

0.04

0.03

0.02

0.01

0.01

0.02

0.03

0.04

X1

(b) Plunge chaotic phase portrait with = 0.1[rad], h[m]

a = 0.4
and = h = 0,U = 25 m
s

The behaviour shown in Fig.1.14a and 1.14b given by a chaotic vector eld for the values of parameters
indicated, and it will be shown in chapter 3 how a MCS controller can recover from chaos.
Setting the initial conditions so to have as solution to the system an LCO behaviour, and because of its
dependance on the parameters as shown in equation 1.12, it is possible to do a bifurcation analysis . Before
showing the results of the bifurcation analysis we qualitatively show on a 3d plot what happens for dierent
values of U and a as depicted in the following gures because it gives a quick glance of how the LCO grow with
respect to velocity:
2 notice

that chaos is possible only for systems of order n > 3, this is a consequence of the Poincar-Bendixson theorem

19

Model and system analysis

600

400

x4 = [degrees/s]

200

200

400

600

800
25
20
40

15
10

35
5

30
0
25

5
20

10
15

x2 = [degrees]

15

U [m/s]

Figure 1.14: Open loop responses on AoA for the system with = 0.1[rad], h = 0.1[m] and = h = 0 and U = 15, 17, 20, 19, 40,
a = 0.4

1.5

x3 = h[m/s]

0.5

0.5

1.5
0.04
0.02
40

0
35

0.02
30

0.04
25

0.06
0.08

x1 = h[m]

20
0.1

15

U [m/s]

Figure 1.15: Open loop responses on plunge for the system with = 0.1[rad], h = 0.1[m] and = h = 0 and U = 15, 17, 20, 19, 40,
a = 0.4

20

1.3. State space representation and system open loop analysis

Similarly as what has been done for U , is done for various values of a at a settled velocity of U = 25.

6
4
2

x3

0
2
4
6
8
0.3
0.25
0.2
0.15
0.1
0.05

0.1
0.2

0.3
0.4

0.05

0.5

0.1

0.6
0.15

x1

0.7
0.8
a

Figure 1.16: Open loop responses on AoA for the system with = 0.1[rad], h = 0.1[m] and = h = 0 and a = 0.1, 0.3, 0.4, 0.6

6
4
2

x3

0
2
4
6
8
0.25
0.2
0.15
0.1
0.05
0
0.05
0.1
0.15
x1

0.8

0.7

0.5

0.6

0.4

0.3

0.2

0.1

Figure 1.17: Open loop responses on AoA for the system with = 0.1[rad], h = 0.1[m] and = h = 0 and a = 0.1, 0.3, 0.4, 0.6

As stated earlier a bifurcation analysis is performed for the LCOs of the system. The bifurcation for the
LCOs is a Hopf bifurcation starting for a = 0.9; no bifurcations for airstream velocity are found. The result
of the continuation calculations is a family of LCOs shown in the following gures.

21

Model and system analysis

1
2

x4

x4

4
6
5
8
0.1

0.05

0.05
x

0.1

0.15

0.2

0.2

0.1

0.1
x2

(a) Family of LCOs for AoA, U = 25 m

s

0.2

0.3

(b) Bigger family of LCOs for AoA, U = 25 m

s

0.3

0.2

0.1

0.1

0.2

0.3

20

15

10

x1

x 10

(c) Family of LCOs for plunge, U =

25 m
s

0.6

0.4

0.2

X3

0.2

0.4

0.6

0.8
0.06

0.05

0.04

0.03

0.02

0.01
X1

0.01

(d) Bigger family of LCOs for plunge, U = 25 m

s

22

0.02

0.03

0.04

1.3. State space representation and system open loop analysis

0.1

0.2

0.2

0.3

0.3

0.4

0.4

0.5

0.5
a

0.1

0.3

0.6

0.6

0.25
0.7

0.7

0.2
0.15

0.8

0.8

0.1

0.9

0.9

0.05
1

1
1

0.05

6
4

0.5

0.1

2
0

0.15

0
x

4
6

0.5

0.2

1
x3

x4

0.2

0.15

0.1

0.05

0.05

0.1

0.15

0.2

x1

(a) 3d visualization of a family of LCOs for AoA, U = 25 m

(b) 3d visualization of a family of LCOs for plunge, U =
s

25 m
s

The analysis shown for the system is only numerical and qualitative. It is although possible even if highly
complex, to show the behaviour of the system use a Lyapunov approach. Being the system mechanical, as a
Lyapunov function it is possible to use the energy of the system; being the system of the 4th order the candidate
Lyapunov function is the following:
Z x2
1
K2,2 ()d > 0
V = ((x3 , x4 )M (x3 , x4)T + K1,1 x21 + x1 x2 K1,2 ) +
2
0
Where M is the mass matrix, Ki,j are the elements of the elastic matrix as in equation 1.7. The calculations of
the time derivative can be carried out, but from what has been shown in the previous gures it is not possible
to show the V < 0; what can be shown is the presence of an invariant region, so to then use LaSalles invariance
principle and thus show analytically the presence of LCO and chaos. Although this is a possible approach the
calculations are particulary dicult, especially to show the invariant regions.
A more viable strategy, given the presence of the LCO behaviour is the use a Poincar Map and Poincar
sections, which is a typical approach in the case of non planar systems.

23

Chapter 2

Input Output Feedback linearization

In this chapter a non-linear controller based on Feedback linearization (FBL) is developed for the systems
introduced in equation 1.12. The main idea of this control technique is to build a non-linear controller so to
cancel the non-linear dynamic terms of the system. After the linearizing law is built a linear controller on a
second control loop is synthesized with classic methods as in [10] or, for instance, an optimal LQ controller can
be developed.
The control objective is to build a possibly global FBL controller, either with partial FBL or total FBL, to
suppress the LCO shown in the previous chapter. Indeed a FBL presents with respect to linearization the
advantage of being global.
There are some downsides in FBL, the main one is that it is not always possible and trivial to fully linearize a
system, indeed sometimes only partial FBL is possible and in this case hidden dynamics are to be investigated
for the stability of the entire system. The second downside, is that the linearizing control law may (and often
does) use parameters that are not always known; if the system parameters change the FBL law may not yield
the mismatch between the assigned parameter in the law and the actual value of the parameter thus may lead
the system to instability. The last downside concerns structural robustness, indeed non modeled dynamics can
aect the systems performance.
Input output FBL (IO FBL) strategy is derived by using partial linearization and thus analysing hidden
dynamics for the system with one control surface. Partial FBL will be applied rst to the AoA control and then
to the plunge dynamics.

2.1

Pitch FBL control

To derive the control law with IO FBL it is necessary to dene an output function y = h(x). In the case of
pitch control the most reasonable variable to considered as output is the AoA since it is also is the easiest to
measure. So another equation is to be added to the system dened in equation 1.12:
y = h(x) = x2
In order to accomplish partial FBL the rst step is to calculate1 the relative degree r of the system. The relative
degree of the system is calculated as in [1]:
h
g(x) = 0
x
h
Lf (h) =
f (x) = x4
x
Lg (h) =

Lg (Lf (h)) = g4 6= 0
1 notice the following calculations the Lie derivative L (h) is calculated because it will be used as both for finding the relative
f
degree r, and for the state transformation as shown in [5]

25

Input Output Feedback linearization

Showing that the relative degree of the system is r = 2 allowing to linearize two equations leaving the other
n r as hidden dynamics. In order to accomplish partial FBL a state transformation is carried out as follows:
x 7 z
z1 = h(x) = x2
z2 = Lf (h) = x4
z3 = x1
z4 = g3 x4 + g4 x3

(2.1)

The criterion in choosing the state transforation is shown in [5]. It is to note that z3 and z4 could have been
chosen dierently but a dierent choice would not have had the benet of having Lg (zi ) = 0, 3 i 4(the
n r hidden dynamics equations ) thus not allowing any eect of the input towards the variables z3 and z4 and
simplifying the hidden dynamics stability analysis. The picked choice for the SS transformation guarantees the
transformation to be invertible.
The new SS representation is the following, again in ane form:
z = f,x (z) + g(z)
where:

f,x (z) =

z2
(k4 + q(z1 ))z1 + (c3 gg34 c2 )z2 k3 z3
1
g4 (z4 + g3 z2 )

c3
g4 z4

((g k g k ) + g q(z ) g p(z ))z + (c g32 + c g c g c g )z + (g k g k )z + ( c1 g3 c )z

4 3
1 3
2 4 2
3 3
4 1 3
1 4
3 4
4 2
3
1
4
1
1
3 g4
g4

g
4
g(z) =

Partial FBL can be achieved by selecting a control law where the non linear dynamics are compensated2 . Thus
by choosing:
(k4 + q(z1 ))z1 + (c3 gg43 c2 )z2 k3 z3 gc34 z4 + v
=
g4
the closed loop system is dened as in equation 2.2:

z2
z

z

v
2
=
1

(g
z
z

3
2 + z4 )
3

g4

g32

z4
((g3 k4 g4 k2 ) + g3 q(z1 ) g4 p(z1 ))z1 + (c3 g4 + c4 g3 c1 g3 c2 g4 )z2 + (g3 k3 g4 k1 )z3 + ( c1gg4 3 c1 )z4
(2.2)
Some considerations are to be made on the closed loop system in 2.2. The rst is that a linear controller v is
to be built for stabilizing the dynamics of z1 and z2 ; secondly the dynamics of z3 and z4 are to be carefully
analysed for stability because there is no direct control on them and more importantly, if not linear, they can
be subject to bifurcations.
The linear controller v is rst build as an LQR controller with a feed forward action, later on a derivative action
will be used as well moving so to a PD controller. The LQR controller is build using a full information scheme,
this is reasonable even if it is not possible to measure all the state vector[z1 z2 ]T , indeed it is still realistic to
measure AoA (z1 ) and estimating the pitch angular velocity (z2 )is not dicult e.g. using a Kalman lter.
2 this

26

in the aerospace industry is know as linear dynamic inversion

2.1. Pitch FBL control

The LQ controller is built by solving the LQ problem in the MATLAB environment, using the Brysons rule to
dene the weight matrixes R and Q. The resulting gain values for the controller are the following:
kz1 = 70.7107, kz2 = 13.8355
The controller scheme design in the Simulink/Matlab environment is shown depicted in the following image:

Figure 2.1: SS graph for AoA partial FBL with = 0.1[rad], h = 0.01[m] and = h = 0

In Fig.2.1 it possible to see the LQ controller3, and the subsystem consisting of the SS transformation and the
compensator for the non-linear dynamics. It should be clear that no transformation is applied to the reference
signal; since the FBL is partial and the transformation -and thus only variables z1 and z2 are of interest- consists
in no more than a simple exchange of position. The full SIMULINK Block diagram is plotted in g.
betaFBLalpha
To Workspace2
t
Clock

To Workspace3

R2D
Radians
to Degrees3

Scope5

Scope4

pfblh
To Workspace4

R2D
Radians
to Degrees2

Scope3

pfblalphatrack
To Workspace

Signal 1

Signal Builder

1/(s/25+1)
Filtro riferimento

reference
z_1,z_2

R2D
Beta

beta

Scope2

Radians
to Degrees1

Fsb

FBL Controller + Transformation

R2D
Plant

Ground

Radians
to Degrees

Scope1

pfblalpha
To Workspace6

pfblalphap
To Workspace1

pfblhp
To Workspace5

Figure 2.2: FBL control scheme for AoA

2.1.1

Simulations

Some simulations for the system are reported in the following gures both for and h. The parameters used
are U = 15 m
s , a = 0.4.

3 not

the feedforward action which is outside of this block.

27

Input Output Feedback linearization

40

35

30

z2=x4= d/dt [deg/s]

25

20

15

10

z1=x2= [deg]

Figure 2.3: Phase Plane of the controlled AoA dynamics with the use of a LQR controller

8
Reference

z1=x2= [deg]

5
Time [seconds]

Figure 2.4: Step response of the controlled AoA dynamics with the use of a LQR controller

28

10

2.1. Pitch FBL control

40

35

30

z2=x4=d/dt [degree/s]

25

20

15

10

5
Time [seconds]

10

5
Time [seconds]

10

[degrees]

10

12

14

Figure 2.5: Time response and control input for = z1 = x2 and = z2 = x4

29

Input Output Feedback linearization

The control goal, which is to suppress LCO and possibly permit tracking of a reference signal, is achieved
on the interested variables. Before discussing the stability of the hidden dynamics it is interesting to see a
qualitative behaviour of the variables [z3 z4 ]. The following gure shows a phase plane of the plunge and its
velocity. Clearly being the relative degree r = 2 there is no direct control over it; and also having chosen the
SS transformation as shown 2.1 allows to have no control input in the hidden dynamics equations.
0.2
0.15
0.1

z4=x3= d/dt h[m/s]

0.05
0
0.05
0.1
0.15
0.2
0.25
0.3
0.03

0.025

0.02

0.015

0.01
z3=x1= h[m]

0.005

0.005

0.01

Figure 2.6: SS graph for the plunge after partial FBL on the AoA. It is clear that the variable is indirectly influenced

Another tracking examples is shown in Fig. 2.7, after appropriate ltering of the reference. In this case the
plunge phase plane also is of interest because it can be seen that not only it is qualitatively stable but it shows
a stable focus as depicted in the following gure and as will be shown further:

10
reference

[degree]

8
6
4
2
0

5
Time [seconds]

10

10

d/dt [degrees/sec]

40

20

20

40

Figure 2.7: Tracking graph and reference signal with null initial conditions

30

2.1. Pitch FBL control

0.01

h[m]

0.01

0.02

0.03

5
Time [seconds]

10

5
Time [seconds]

10

0.2

d/dt h [m/s]

0.1
0
0.1
0.2
0.3

Figure 2.8: Plunge position and speed with null initial conditions when the AoA subject to reference signal as in Fig. 2.7

FBL control is highly model based and it is interesting to show a simple example of what happens when
there is uncertainty on the value of one or parameters. In the following gures a FBL law is built by considering
for the air stream velocity value at UF BL dierent from the real parameter U of the system. Indeed a 1g positive
acceleration is introduced a t = 5.9[s]; the following gure show the time response.

8
Reference

5
Time [seconds]

10

Figure 2.9: Tracking error for give by building FBL control law on different parameters (UF BL , aF BL ) from the systems one
(U,a)

A constant tracking error and an increase of settling time with respect to Fig. 2.7 are present that the LQ
controller is not capable of rejecting. Some more information can be obtained by looking at the plot of :

31

Input Output Feedback linearization

40

35

30

25

d/dt

20

15

10

5
Time [seconds]

10

Figure 2.10: dynamics when subject to non rejected parameter variations

Indeed when using FBL a robust linear controller is essential in order to handle unexpected parameter
variation that are not handled by the non linear dynamic compensating law. As stated earlier, for this purpose
a derivative action is introduced in the controller so to allow to reject some parameter variations. After the
introduction of a PD controller the resulting step response allows a null tracking error as can be seen in g.t
would have been possible also to introduce a LQI scheme or a PI linear controller, but in this occasion a
derivative action is preferred.

8
Reference
FBL with uncertain parameters tracking

[deg]

5
Time[seconds]

Figure 2.11: Null tracking error for subject to parameter variation with a PD controller linear controller

32

10

2.1. Pitch FBL control

2.1.2

Hidden dynamics analysis

The hidden dynamics are to be investigated; in order to accomplish the stability analysis the zero dynamics of
the system are investigated, by setting [z1 z2 ]T = 0 thus leading to the following system:
(

z3
z4

"

0
2,1

1,2
2,2

#(

z3
z4

(2.3)

where some constants of system 2.2 have been renamed as follows:

1,2 =

2.1.2.1

1
,
g4

2,1 = (g3 k3 g4 k1 ),

2,2 =

c1 g 3
c1
g4

Stability analysis

As qualitatively shown in Fig 2.6 there is a stable focus in [z3 z4 ] = (0, 0) that can be easily found solving the
system 2.3 with null solution. The eigenvalues calculated using as parameters the same as previous simulations
a = 0.4 U = 15m/s. The eigenvalues are: [1.3122 + 17.1258j], [1.3122 17.1258i] showing so a stable focus,
in accordance with what has been qualitatively shown in previous Fig.2.6. It is important to notice that the
zero dynamics system is linear, but still parameter dependant; it is necessary so to evaluate the position of the
eigenvalues for dierent values of a and of the velocity U as illustrated in Fig. 2.1.2.1

12
10
8

Eigenvalues

6
4
2
0
2
4
0

6
20
8
1

40
0.9

0.8

0.7

60
0.6

0.5

0.4

0.3

80
0.2

0.1

100

U [ ms ]
a

Figure 2.12: Plot of the eigenvalues with respect to velocity U and a. Color pattern towards red indicates increasing values for the
eigenvalues

Fig.2.1.2.1 shows the how the real part of the eigenvalues is smaller than zero for all velocity values a
a < 0.55 and greater than zero for a > 0.55
Another possibility to investigate the systems stability is to use the Lyapunov equation. This technique is not
followed since the use of the stability analysis using eigenvalues is a more synthetic way to for linear systems.
Secondly being the system parameter dependant the use of eigenvalues well shows how the stability varies with
respect to U and a; indeed using the Lyapunov equation would have requested to analyse a parameter varying
matrix, solution of the equation: AT P + P A < 0 with P = P T > 0.
33

Input Output Feedback linearization

2.2

Plunge FBL control

The FBL control law for plunge is built in a similar way as it was done for the AoA partial FBL control. At
rst an output function is selected:
y = h(x) = x1
After this the relative degree r of the system is calculated:
h
g(x) = 0
x
h
Lf (h) =
f (x) = x3
x
Lg (h) =

Lg (Lf (h)) = g3 6= 0
The relative degree of the system is r = 2 thus allowing a partial FBL on the plunge dynamics. Since r < n
there are 2 hidden dynamics to be investigated.
Following the calculation of r, a SS transformation is introduced as following, again, the criterion shown by [5]
is used:
x 7 z
z1 = h(x) = x1
z2 = Lf (h) = x3
z3 = x2
z4 = g3 x4 + g4 x3

(2.4)

Fortunately also in this case it is possible, and easy, to nd a SS transformation where Lg zi = 0, 2 i 4

making orthogonal the variables not interested by the FBL law to the control input.
The the transformation introduced in 2.4 allows a recast of the systems to the ane form of equation 2.2 where:

z2

g4

k1 z1 (c1 g4 + c2 g3 )z2 ) (k2 + p(z3 ))z3 gc23 z4

f,x (z) =
1

g3 (g4 z2 + z4 )

(g k k g )z + (c + c g42 c g c g )z + (g (k + p(z )) g (k + q(z )))z + ( g4 c c )z

3 3
4 4 2
4 3
3
3 4
3
3
4 4
4 1
3 3 1
1
2 g3
g3 3

3
g(z) =

By selecting an appropriate control law plunge FBL can be achieved. Thus chosen as follows:
=

+k1 z1 + (c1 g4 + c2 gg34 )z2 ) + (k2 + p(z3 ))z3 +

c2
g3 z4

+v

g3

The resulting system equations after the compensator law has been selected as following:

z2
z

z
v
2
=
1

z3

g3 (g4 z2 + z4 )

(g k k g )z + (c + c g42 c g c g )z + (g (k + p(z ) g (k + q(z ))))z + ( g4 c c )z

z4
3 3
4 4 2
4 3
3
3 4
3
3
4 4
4 1
3 3 1
1
2 g3
g3 3
(2.5)
It is now essential to build a linear controller v on an outer control loop so to stabilize the plunge dynamics.
Also in this case,a s what has happened for the IO FBL of the AoA an LQ controller is developed. By solving
the LQ problem and choosing the weight matrixes using the Brysons rule the following gains are obtained:
kz1 = 70.7107, kz2 = 13.8355
34

2.2. Plunge FBL control

In the case of the plunge dynamics no derivative or integral action is used, but it is clear that it can easily bee
implemented.

2.2.1

Simulations

With the linear controller v deployed some simulations can be shown before analysing the zero dynamics of the
system. In the following gures some signicant simulations are shown:

0.1

0.1

d/dt h [m/s]

0.2

0.3

0.4

0.5

0.6

0.7
0.02

0.02

0.04
h [m]

0.06

0.08

0.1

Figure 2.13: Phase plane for a closed loop response for partial FBL on the plunge dynamics, with the following initial conditions
h = 0.1 and = = h = 0

0.1

0.08

h [m]

0.06

0.04

0.02

0.02

5
Time [seconds]

10

Figure 2.14: Time response for partial FBL on the plunge dynamics, with the following initial conditions h = 0.1 and = = h = 0

35

Input Output Feedback linearization

0.1

0.1

d/dt h[m/s]

0.2

0.3

0.4

0.5

0.6

0.7

5
Time [s]

10

Figure 2.15: Time response for partial FBL on the plunge dynamics, with the following initial conditions h = 0.1 and = = h = 0

Clearly the controller stabilises the system for the plunge dynamics for which there is no reason to have
tracking but just a regulation to zero. When dealing with partial FBL a qualitative check to the dynamics with
no control is often interesting as show in the following phase plane gure:

300

200

d/dt [degrees/sec]

100

100

200

300
5

[degrees]

10

15

20

Figure 2.16: Phase plane for and for a closed loop response for partial FBL on the plunge dynamics, with the following initial
conditions h = 0.1 and = = h = 0

2.2.2

Hidden dynamics analysis

To analyse the systems zero dynamics the variables z1 and z2 are set to zero, obtaining so the following
equations:
(
) (
)
z3
1,2 z4
=
(2.6)
z4
(g4 (k2 + p(z3 )) g3 (k4 + q(z3 )))z3 + 2,2 z4
36

2.2. Plunge FBL control

Where 1,2 and 2,2 rename some constants from system 2.2 as follows:
1,2 =

1
,
g3

2,2 =

g4
c2 c4
g3

It is clear that the zero dynamics of the system are a family of non-linear equations depending on parameters
U and a.
2.2.2.1

Lyapunov stability analysis

To analyse the stability of the zero dynamics system equations 2.6 a Lyapunov theory approach is used. The
rst step to simplify the process of nding a Lyapunov function is to recast the system to 2nd order ODE. This
can be achieved as follows:
)
) (
(
1,2 z4
z3
=
=
((g4 mxd b g3 m
z4
d )K (z3 ) + (g4 k2 g3 k4 ))z3 + 2,2 z4
(

1,2 z4
19.94z3 54.57z32 + 2593.6z33 16916.5z34 + 34088.5z35 + 2,2 z4

And thus we can recast the hidden dynamics form as if it is a non linear mass-spring-damper system:
z 2,2 z (z) = 0,

(z) = 4.51z + 9.86z 2 5.87z 3 + 3.82z 4 7.71z 5

We now search for a candidate Lyapunov function. A possibility is to use the energy of the system, which is
the sum of kinetic and potential energy. Thus the candidate Lyapunov function V (z) is selected as follows:
V (z) =

1 2
z +
2

() =

1 2
z + 2.25z 2 3.28z 3 + 1.46z 4 0.774z 5 + 1.28z 6
2

(a)

(b)

Figure 2.17: Lyapunov function for the plunge Zero dynamics. It is clearly continuous, differentiable, null in (0, 0), positive definite
and radially unbounded

and use the input-output form of the zero dynamics we obtain:

Now if we calculate V (z)
z(
z + (z)) = 2,2 z 2
Thus the system is locally asymptotically stable because 2,2 < 0 as can bee seen from Fig.2.18.

37

Input Output Feedback linearization

Figure 2.18: Derivative of the Lyapunov function of Fig.2.18.

A possibility to show that the system is globally stable, is to use the La Salle invariant set theorem and so
use the so called sector conditions as shown in [1]. The sector conditions, given a 2nd order ODE in the form
of y + d(y)
+ ky = 0, are the following:
d(y)y
> 0,

x>0

k(y)y > 0,

x>0

d(0) = k(0) = 0
if the conditions are full lled then the system is globally stable because being the biggest invariant set for which
V = 0 the equilibrium point in (0, 0). In this case as can be seen in the following gures the condition are full
lled thus the zero dynamics are globally stable.
A second possibility to show that the system is globally stable is to use the Babashin Krasovskii theorem;indeed from what can be seen in earlier gures, the condition on radial unboundedness is respects, and
= 2,2 z 2 < 0 we can conclude that the system is globally stable.
having V (z)
2.2.2.2

Bifurcation analysis

For an accurate stability evaluation, since the system in 2.6 is parametric, it is important to conduct a bifurcation
analysis. The rst step is to calculate the equilibrium point of the system. By posing [z3 z4 ] = (0, 0) it is
trivial to obtain [z3 z4 ] = z = (0, 0). The equilibrium point can move from being stable to un stable depending
on the values of parameters U and a.
Following [1], the rst step to evaluate the possibility of bifurcation for the consider system is by using the
bifurcation existence conditions as follows:
= (0, 0)
f (
z , )
f
= (0, 0)
(
x, )
x

(2.7)

Where f is the vectorial function describing the system in 2.6 and z, are respectively the equilibrium point
considered and the corresponding value of the parameter - or parameter vector - for which the bifurcation
occurs. Bifurcations can so occur for this system as shown by the previous calculations. Indeed the following a
numerical evaluation is shown rst with respect to parameter = U 2 and then with respect to a

38

2.2. Plunge FBL control

0.1

0.08

0.06

0.04

z3

0.02

0.02

0.04

0.06

0.08
0

100

200

300

400

500

600

700

800

Figure 2.19: Bifurcation diagram with respect to and with a = 0.4

0.08

0.06

0.04

z3

0.02

0.02

0.04

0.06

0.08

100

200

300

400

500

600

700

800

Figure 2.20: Bifurcation diagram with respect to and with a = 0.5

39

Input Output Feedback linearization

0.06

0.05

0.04

0.03

z3

0.02

0.01

0.01

0.02

0.03

0.04

100

200

300

400

500

600

700

800

700

800

Figure 2.21: Bifurcation diagram with respect to and with a = 0.6

0.05

0.04

0.03

z3

0.02

0.01

0.01

0.02

0.03

100

200

300

400

500

600

Figure 2.22: Bifurcation diagram with respect to and with a = 0.63

It is important to notice that Fig.2.19,2.20,2.21,2.22 show a pitchfork bifurcation but it is not a symmetric
bifurcation. Indeed this sort of bifurcation is also know as an imperfect bifurcation or perturbated bifurcation as
showed in [13] and [4]. In These kind of bifurcation the normal form of the bifurcation, that in the case of a
standard pitchfork is x = x x3 , becomes x = x + x2 x3 , therefor adding a quadratic term. Although
these bifurcations are perturbated it is still possible to show a critical velocity c after which the uncontrolled
plunge dynamics equilibrium is unstable.
Bifurcation diagrams are show with respect to a, varying as shown in the Fig.2.23 and 2.24 :

40

2.2. Plunge FBL control

0.08

0.06

0.04

z3

0.02

0.02

0.04

0.06

0.08
1

0.9

0.8

0.7

0.6

0.5
a

0.4

0.3

0.2

0.1

0.2

0.1

Figure 2.23: Bifurcation diagram with respect to a and with = 115

0.15

0.1

z3

0.05

0.05

0.1

0.9

0.8

0.7

0.6

0.5
a

0.4

0.3

Figure 2.24: Bifurcation diagram with respect to a and with = 115

41

Chapter 3

MRAC Minimum Controller Synthesis

control for flutter
In this chapter an MRAC adaptive controller will be applied to the plant dened in chapter 1 using the Minimum
Controller Synthesis algorithm (MCS)1 . We will show briey how the MCS algorithm works before showing the
results, and the extensions of the MCS and simulations.

3.1

MCS algorithm

The MCS strategy is based upon an extension of Landaus MRAC approach [6] as shown in [11]. The MCS
control strategy aims to track asymptotically a reference model as dened in 3.3 trough a gain adaptation
law. One of the most important characteristics of the control strategy is that there is no assumption on the
knowledge of the plant parameters. The only assumption that is held is that the plant structure is known and
fully controllable in a canonical form like follows:
x = Ax(t) + Bu(t)

(3.1)

where x(t) Rn ,A(t) nn ,u(t) and vector B n1 are:

0
0
1
0

0
1

0
0
0

A=
..
..
..
..
B = ..
..
.
.
.
.
.
.
b
a1 a2 a3 an

(3.2)

Given the plant as in 3.1, it is necessary to dene a reference model as follows2 :

x m = Am xm (t) + Bm r(t)
where
Am

0
=
..
.
am1

1
0
..
.
am2

0
1
..
.
am3

The control law used in the MCS strategy is the following:

..
.

0
0

0
0

..
Bm = ..
.
.
1
amn

uMCS (t) = K(t)x(t) + KR (t)r(t)

1 from
2 note

now on the MCS algorithm will be referred to as if it a control strategy with a slight name abuse
that the reference model is supposed to be stable, as so, the matrix Am is a Hurwitz matrix

43

(3.3)

(3.4)

(3.5)

MRAC Minimum Controller Synthesis control for flutter

where:
K(t) =

t
T

ye ( )x ( )d + ye (t)x (t)

KR (t) =

ye ( )r( )d + ye (t)r(t)

(3.6)

K(0) = 0, KR (0) = 0, xe (t) = xm (t) x(t)

In the equation 3.6 the terms ye and is dened as:
ye (t) = Ce xe (t),
and P is the solution of the Lyapunov equation

h
Ce = 0

P Am + ATm P = Q

i
0 ... 1 P

P > 0, P = P T

With the control law dened in 3.5 the error system shown in gure 3.1, is asymptotically stable, as proven
in [1]

Figure 3.1: MCS scheme

The MCS controller can be applied to non linear plants, as it has been shown in [12] that MCS control strategy
rejects non linear disturbances3 and vanishing disturbances towards the input direction.

3.1.1

MCS extensions

Some extension of the MCS control strategy have been proposed in literature. Part of the extensions focus on
conjugating optimal control with the MCS control strategy, while others aim to modify the control law so obtain
specic proprieties of the controlled system.
3.1.1.1

MCS-LQ

In this extension of the MCS strategy the reference model is an optimally controlled linear model. The reference
model, that in applications is typically derived from the non linear model of the plant, is controlled with an
optimal LQ controller so to reach desired proprieties of optimality. This scheme, show in gure 3.2 allows to
increase the LQ robustness because the MCS strategy compensates the mismatch between plant and the LQ
optimal trajectory.

3 note that in this propriety well fits with the use of the MCS strategy that is used in this report. Indeed in aeronautics often
not only parameters are unknown but non linear effects cannot be correctly modeled due to order reduction and other factors

44

3.1. MCS algorithm

Figure 3.2: LQMCS scheme

3.1.1.2

MCSI

This extension of the MCS strategy modies the control law by inserting an integral action (MCSI). This
extension in necessary in those cases in which it is mandatory to have a null tracking error. The MCS control
law is modied as in equation 3.7:
uMCSI (t) = K(t)x(t) + KR (t)r(t) + KI (t)xI (t)
In equation 3.7 the terms KI (t) and xI (t) are dened:
Z
Z t
xI (t) = (r y)
KI (t) =
ye ( )xI ( )d + ye (t)xI (t),

(3.7)

KI (0) = 0

(3.8)

The stability of the MCSI strategy has been proven in [1].

3.1.1.3

EMCS

EMCS stands for Extended MCS and it is a modied control of the standard MCS law so to reject generic
disturbance.
The MCS law presented in 3.5 is modied by adding a commuting control action:
uEMCS (t) = K(t)x(t) + KR (t)r(t) + N sgn(ye )

(3.9)

where the value of N must respect the following propriety as shown in [1]:
1
max{|d|}
(3.10)
b
It is important to observe that in control law 3.9 it is necessary to have a measurement, or at least an estimation,
of the disturbance d. The main advantage of using the control law in 3.9 is so have a more robust controller.
This can be simply understood by recalling the main results of commuting controllers such a Sliding controllers.
N>

3.1.1.4

NEMCS

Based upon 3.9, and based upon the fact the value N is set by a measurement of the disturbance amplitude,
the NEMCS introduces a gain varying law for the commuting action.
uN EMCS (t) = K(t)x(t) + KR (t)r(t) + Kn (t)sgn(ye )

(3.11)

In equation 3.11 Kn (t) is dened as follows:

Kn (t) =

ye (t)

(3.12)

the proof of stability for this controller, with some interesting experimental results, is given in [7]
45

MRAC Minimum Controller Synthesis control for flutter

3.1.1.5

LQ-NEMCSI

The LQ-NEMCSI an extension of the LQ-MCS and NEMCS control strategy where the control laws are modied
so to have integral action and good rejection capabilities following so an optimal trajectory from the linear model.
The control law is dened as follows:
uN EMCSI (t) = K(t)x(t) + KR (t)r(t) + Kn (t)sgn(ye ) + KI (t)xI (t)

(3.13)

It has been show in [7] to be asymptotically stable.

3.2

MCS control synthesis and simulations

In this section MCS and some extensions of MCS control strategy are deployed to control the plant dened in
1.7. The main goal is the suppression of LCO and tracking of a given reference signal.
The system dened in 1.12 is dened with x 4 and with and does not result in the ane canonical
form. This poses a rst problem in applying the MCS strategy. Indeed to control such a system a possible
solution is to add a second control surface and dening two MCS controllers each one working on a separate
ane system and rejecting the eects of the remaining part of the system. Another possible solution, marly
qualitative, is to use only on control input for allow tracking on AoA and show that the plunge dynamics is
stable. In this report two dierent controllers are deployed one for AoA and one for plunge dynamics.
As a rst step we can rewrite the system dened in 1.12 f and g functions in so to separate clearly what is seen
as a disturbance from the controller:

x3
0

x4
0
(3.14)
=
f,x (x) =
,
g(x)
2.82mx b

)x2 c1 x3 c2 x4 + d3 (x, t)

k1 x1 (k2
g3

g4
k3 x1 (k4 + 2.82m
d )x2 c3 x3 c4 x4 + d4 (x, t)
where d3 (x, t) and d4 (x, t) as the following:

d3 (x, t) = p(x2 )

2.82mx b
,
d

d4 (x, t) =

2.82m
+ q(x2 )
d

The following step is to dene a reference model for both the plunge and AoA dynamics. This is accomplished
by selecting Am and Bm as the following matrixes:
"
#
" #
0
1
0
Am =
, Bm =
(3.15)
1000 75
1
The step response of this model is shown in Fig.3.3 and has no overshooting and a settling time of approximately
0.3[seconds].

46

3.2. MCS control synthesis and simulations

1
Reference
Position

Amplitude [deg] [m]

0.8

0.6

0.4

0.2

0.1

0.2

0.3

0.4

0.5
Time [s]

0.6

0.7

0.8

0.9

Figure 3.3: Reference model step response

Under the assumption of a single control surface and thus controlling only the AoA dynamics4 the closed
loop equations can be given for a MCS controller as in 3.16:

x 1
x3

x4
2
(3.16)
=
b

)x2 c1 x3 c2 x4 + d3 (x, t) + g3 MCS

x 3
k1 x1 (k2 2.82mx

k3 x1 (k4 + 2.82m
x 4
d )x2 c3 x3 c4 x4 + d4 (x, t) + g4 MCS
Similarly by changing the control law for it is possible to write the closed loop equations for the various MCS
extensions.
If using two control surfaces by recalling system dened in 1.10 it is possible, with some manipulation, to redene
in a SS representation the closed loop equations as in equation 3.17:

x3
x 1

x4
2
=
(3.17)
2.82mx
b

k
x

(k

+
g

)x

c
x

c
x
+
d
(x,
t)
+
g

1
1
2
3
23
2
2
1
3
2
4
3
13
1

M
CS
M
CS
d

x 4
k3 x1 (k4 + 2.82m
d )x2 c3 x3 c4 x4 + d4 (x, t) + g14 1M CS + g24 2M CS

As stated previously since the system in not in canonical form two dierent controllers are deployed one for
controlling the plunge dynamics and one for controlling the AoA as can be seen in Fig.?? thus the two control
surface model is used. Clearly the is no necessity for the plunge dynamics to have tacking capabilities and the
plunge controller has no feed forward, and the control surface used is smaller than the one used for the AoA
dynamics thus it is just a stabilising controller. Also, in a realistic application there is no real need to have
plunge control that introduces a cost growth, due to the introduction of a seconds control surface; indeed in
the aerospace industry the plunge dynamics typically is not controlled but some action are taken so to reduce
or cancel undesired behaviours. From now on only simulations for the AoA are shown, but the reader should
keep in mind that all the simulation have on the plunge dynamical a gently tuned stabilising controller. Only
one simulation is displayed in g.3.4a with the relative feedback gains n g.3.4b.

4 it is no difficult to imagine that this is the main goal for control, plunge can be controlled with structural stiffness or wight
distribution

47

MRAC Minimum Controller Synthesis control for flutter

R2D
Radians
to Degrees2
R2D
Radians
to Degrees
rif
beta1

beta1
x

MCS h
beta2

R2D
Plant

Radians
to Degrees1

rif

Filtered_rif

Beta1

Riferimento filtrato

R2D

Radians
to Degrees3

MCS
alpha

0.02
30
0
20
0.02

10
Feedback gains

h [m]

0.04

0.06

0.08
10
0.1
20
0.12

0.14

30
0

0.1

0.2

0.3

0.4

0.5
0.6
Time [seconds]

0.7

0.8

0.9

(a) Controlled plunge dynamics with MCS

3.2.0.6

0.1

0.2

0.3

0.4
0.5
Time [seconds]

0.6

0.7

0.8

0.9

(b) Feedback gains for controlled plunge dynamics

Simulations

Some simulations are shown using the equations dened in the previous sections. At the beginning of this
chapter it is stated that no assumption is made on the numerical values of the plant parameters; this on one
side means that the MCS controller works as an identier other than as a controller - as can be seen in [1], and
on the other side means that in order to allow a correct identication of the plant parameters a transient is
necessary so to allow the gains to adapt. In this report this phase will be shown by using long simulations but,
as stated in [7], a more industrial approach would be to use as parameters from linear models of the system as
a rough rst estimation. This, that might seems a minor aspect for the use of the MCS control, turns out to
be a limitation. Indeed, as will be shown further, if no knowledge of the plant parameters is assumed the gain
adaptation law will start with a peak that will be reected on the control input. This can result in unexpected
behaviours on the real plant that can lead to instability, LCO, or chaos because of un modelled dynamics. Due
to these reasons the reference signal for the AoA dynamics is periodic square wave at a frequency of 0.2[Hz].
A single period of the reference signal is shown in Fig3.4. If necessary a rst order lter is used so to slow
down the reference signal with a bandwidth of approximately 10 rad
s . If other reference signals are used it will
be clearly stated.

48

3.2. MCS control synthesis and simulations

0.16
Reference
0.14

0.12

Amplitude [deg] [m]

0.1

0.08

0.06

0.04

0.02

0
0

0.5

1.5

2.5
Time [s]

3.5

4.5

Figure 3.4: Reference signal period

Simulations are shown on wide time range so to allow the gains to settle and there are performed on the
basic MCS, some extension, and some possible advanced use cases and problems are also addressed.
The rst simulations are conducted under the assumption of a two control surfaces, only the AoA dynamics,
since a simply stabilising controller is on the plunge dynamics, as stated previously.
3.2.0.7

MCS controller

An MCS controller, as show in Fig. ?? , is so deployed using the following parameters = 15000, =
the results are show in the following gures:

10

and

x = Ax+Bu
y = Cx+Du
StateSpace

ye
K0

Matrix
Multiply

K*u
Ce

Radians
R2D to Degrees
r
Kr

1
rif

ye

Matrix
Multiply

Beta1
1

r
Ki*int
ye

ye

KN(t)*sgn(ye)

49

MRAC Minimum Controller Synthesis control for flutter

12
Reference
Angle of attack
10

Angle of attack [deg]

50

100

150

Time [s]

Figure 3.5: Reference signal and output for a MCS controller on the AoA dynamics

20

[deg]

20

40

60

80

100

50

100

150

Time [s]

Figure 3.6: Control signal for a MCS controller on the AoA dynamics

In Fig.3.5 it can be seen how the controller adapts the gains after a transient. It is also important to notice
that there happens to be no null tracking error because of the lack of an integral action

50

3.2. MCS control synthesis and simulations

8
Error
6

Error [deg]

50

100

150

Time [s]

Figure 3.7: Tracking error for a MCS controller on the AoA dynamics

In Fig.3.7 the tracking error is shown; during the transient in which the gains are adapting the error is
consistent but gradually decreases as the controller adapts the gain. The tracking error does not go to zero
specially in the rise phase of the signal, this problem will be solved, at least in stationary conditions by the use
of a MCSI controller.
When using an adaptive control scheme, like the MCS, one of the most important things to check is the values
of the gains. This is crucial in the use of a MCS control scheme since, as for instance in the case of saturations
o persistent disturbances, the gain can assume very high values leading so the system to instability. In this case
so the feed forward and feedback gains are show in Fig.3.8 and Fig3.9.

0.5

KR

1.5

2.5

3.5

50

100

150

Time [s]

Figure 3.8: Feed Forward gains for the single MCS controller on the AoA dynamics

51

MRAC Minimum Controller Synthesis control for flutter

50

100

150

Time [s]

Figure 3.9: Feed back gains for the single MCS controller on the AoA dynamics

In Fig.3.9 it is important to notice two details: the rst regards the initial peak of the gains this often in
practical applications has to be saturated since can cause big initial control actions, the second aspect regards
the gains that do not settle at a xed value. What really happens is that the gains do settle at a xed value but
over a much bigger period of time as illustrated in Fig.3.10. The reason for this behaviour is to be researched int
the persistent disturbances the controller is faced to contrast due not only to the uncontrolled plunge dynamics
but also to the non linearities. As will be shown later other control actions will allow to reduce the settling time
for the gains, and their values.

15

10

10

500

1000

1500

2000

2500
Time [s]

3000

3500

4000

4500

5000

Figure 3.10: Feed back gains for the single MCS controller on the AoA dynamics

Before showing the results obtained with a MCSI controller a closer look to the control signal and relative
feedback gain is of interest so to show what happens on a short period of time. Indeed the gain grows briey
for the single maneuver but the reassest on a lower value. The reason for such a behaviour is because the
MCS control action does not only have a integral law for the gain adaptation but also a faster proportional law
52

3.2. MCS control synthesis and simulations

weighed by the parameter .

10
1

5
0

0.5

[deg]

0
10
15

0.5

20
1
25
1.5

30
35

2
77

78

79

80
Time [s]

81

82

83

84

76

77

78

79

80
81
Time [s]

82

83

84

85

Figure 3.11: Control signal and relative Feedback gain for a single maneuver

3.2.0.8

MCSI controller

An implementation of an MCSI increases the performance of the closed loop system as show in the following
gures.

(x2) [deg]

10

20

30

40

50
Time [s]

60

70

80

90

100

Figure 3.12: Reference signal and output for a single MCSI controller on the AoA dynamics

53

MRAC Minimum Controller Synthesis control for flutter

Tracking error [deg]

10

20

30

40

50
Time [s]

60

70

80

90

100

Figure 3.13: Tracking error for a single MCSI controller on the AoA dynamics

Dierently from what happens in Fig.3.13 the tracking error is null specially after the transients for the rise
time of the reference signal.

40

20

[deg]

20

40

60

80

100

120

10

20

30

40

50
Time [s]

60

70

80

90

100

Figure 3.14: Control signal for a single MCSI controller on the AoA dynamics

In Fig. 3.14 the control signal is shown; the major dierence with the Fig.3.6 results in smaller peak values
of the control signal but a closer look shows in Fig.3.18a, also a smoother control action. The feedback and feed
forward gains for the MCSI control settle in a much shorter time with respect to what happens in the MCS
controller (a good assessment can be seen from Fig. 3.10 after more or less 3000[s] ). This is due to the integral
action show in Fig.3.17 that helps a faster stabilisation. The gains for feed forward, feedback and integral action
can be seen respectively in Fig.3.16, 3.15 and Fig.3.17.

54

3.2. MCS control synthesis and simulations

10

15

20

25

30
Time [s]

35

40

45

50

55

Figure 3.15: Feed back gain for the single MCSI controller on the AoA dynamics

0.5

KR

1.5

2.5

10

20

30

40

50
Time [s]

60

70

80

90

100

Figure 3.16: Feed Forward gains for the single MCSI controller on the AoA dynamics

55

MRAC Minimum Controller Synthesis control for flutter

0.8

0.6

KI

0.4

0.2

0.2

0.4
10

20

30

40

50
Time [s]

60

70

80

90

Figure 3.17: Integral gain KI for the single MCSI controller on the AoA dynamics

Before showing other controller implementations a comparison is helpful to show the increase of performance
between the MCS ad MCSI control strategy as can be seen in the following gures:

56

3.2. MCS control synthesis and simulations

8.8

8.6

8.4
6

(x2) [deg]

(x2) [deg]

8.2
5

7.8

7.6

7.4

7.2

MCSI
MCS
Reference

0
68.5

69

69.5

70

70.5
Time [s]

71

71.5

72

71

MCS
MCSI

MCSI
MCS
Reference

7
71.2

71.4

71.6

71.8
Time [s]

72

72.2

72.4

72.6

MCSI
MCS

20

15
4

5
Control signals [deg]

Traking errors MCS VS MCSI [deg]

10

10

15
4
20

25

6
69

70

71

72
Time [s]

73

74

75

73.5

74

74.5

75

75.5

76
Time [s]

76.5

77

77.5

78

78.5

Figure 3.18: MCS vs MCSI controller on AoA dynamics

3.2.0.9

EMCSI controller

An EMCSI controller is implemented in this section and a rst comparison result is shown with respect to the
MCSI controller in the rst seconds of simulation.
The dimensioning of the parameter N has been carried out with in a qualitative manner as follows. In equation
3.14, all values of variables x1 and x3 have been substituted with what is a reasonable physical value for them.
The function d4 has been evaluated a 1.5[rad] - which is a high value for the angle of attack). Then by following
the rules showed in 3.10 N results approximately 15000. Thus by leaving the parameters and the same
as what has been done with the MCS and MCSI simulations from Fig.3.19, it is possible to see that time to
settle good performance is much lower with respect to the MCSI. The reason for such behaviour is to seek in
the commuting action that introduced by the EMCSI controller, that as stated before make the system more
robust.

57

MRAC Minimum Controller Synthesis control for flutter

MCSI
Reference
EMCSI

(x2) [deg]

0.5

1.5

2.5
Time [s]

3.5

4.5

Figure 3.19: Reference signal and output for a single EMCSI controller on the AoA dynamics

Similarly for what has been done for the MCS and MCSI controller the gains are plotted so to show that
the not go grow indenitely.

0.01

0.01

0.02

0.03

0.04

0.05

50

100

150

Time [s]

Figure 3.20: EMCSI Feedback gains for AoA dynamics with one control surface

58

3.2. MCS control synthesis and simulations

0.01

0.5

0.4
0
0.3
0.01

0.2

0.1

KR

Kn sgn(ye)

0.02

0.03
0.1

0.2

0.04

0.3
0.05
0.4

0.06

50

100

0.5

150

50

100

Time [s]

150

Time [s]

Figure 3.21: EMCSI gains for AoA dynamics with one control surface

The down side of this control scheme is to be searched in the commuting action introduced on the actuator
and show in Fig.3.22.

30

20

[deg]

10

10

20

30
32

34

36

38
Time [s]

40

42

44

Figure 3.22: Control signal for a single EMCSI controller on the AoA dynamics

A nal simulation with respect to the MCSI controller is shown in Fig.3.23

59

MRAC Minimum Controller Synthesis control for flutter

Reference
EMCSI
MCSI

0.03

0.025

0.02

(x2) [deg]

0.015

0.01

0.005

0.005

0.01
78.9

79

79.1

79.2

79.3
Time [s]

79.4

79.5

79.6

79.7

Figure 3.23: Controller tracking comparison between MCSI and EMCSI controller on the AoA dynamics after a longer simulation

3.2.0.10

NEMCSI controller

For this controller only major results are shown as a comparison with the EMCS and the values of the gains.
In Fig.3.24 it is possible to see after the gain have reached a stable value a performance increase.

NEMCSI
Reference
EMCSI

NEMCSI
Reference
EMCSI

6
8.5
(x2) [deg]

(x2) [deg]

3
8
2

0
101.5

102

102.5

103
Time [s]

103.5

104

104.5

7.5

101.2

101.3

101.4

101.5

101.6
101.7
Time [s]

101.8

101.9

Figure 3.24: Output performance comparison between NEMCSI and EMCSI

The gains are reported, with exception for KN reported in Fig.3.26, are reported in Fig. 3.25.

60

102

3.2. MCS control synthesis and simulations

0.5

0.5

1.5
Feedback Gain K
Feedback Gain K
Feedforward gains KR
10

20

30

40

50

60

70

80

Figure 3.25: NEMCSI controller gains on the AoA dynamics after a gain assessment, with exception of KN reported in Fig.3.26

x 10

KN

10

20

30

40
Time [s]

50

60

70

Figure 3.26: NEMCSI controller KN gain on the AoA dynamics after a gain assessment

3.2.0.11

LQ-NEMCSI controller

This extension of the MCS strategy uses an LQ controlled linear model as a reference model for the plant
and using the control laws seen for he NEMCI. An LQ model is o synthesised by solving the LQ problem on
the reference model as in equation 3.3; the selection of the weight matrixes is done using the Bryson rule.
The simulation results obtained, and showed in comparison to the other MCS extensions, are show on in the
following gures.

61

MRAC Minimum Controller Synthesis control for flutter

9
Reference
LQNEMCSI
NEMCSI
EMCSI

(x2) [deg]

100

100.5

101

101.5

102

102.5
Time [s]

103

103.5

104

104.5

105

Figure 3.27: LQ-NEMCSI controller performance

10
Feedforward gain
Feedback gain
Feedback gain

Integral gain KI(t)

4
1
2
0
Gains

4
2
6
3

10

4
10

20

30

40

50
Time [s]

60

70

80

90

10

20

30

40

50

60

70

80

Figure 3.28: Gain plot for LQ-NEMCSI

3.2.0.12

Gain-locking

It is important to show that the MCS control strategy is not free of problems. One of the major ones is to be
searched in the innite value that the controller gains can reach. A very simple complication of the system allows
to show how this can severely aect the performance of the controller. A saturation can be introduced to model
the actuator limitation for high values of the angles, the behaviour is shown in the following gure. Saturation
on the control input is introduced and as can be seen in Fig.3.29a the values of the gains grows indenitely.
This happens because the controller reads a null tracking error that cannot compensate fully because of the
saturation o the actuators; the result of this a fast growth of the gains. To solve this issue, that is also known
as gain drifting, a gain locking criterion has to be introduced. As for shown example a possible solution is to
lock the gains after that the tracking error is below 1% and 1[s] has passed after the assigned settling time.

62

3.2. MCS control synthesis and simulations

20
Reference

15

7
10
6

(x2) [deg]

[deg]

5
3

10

15

20

0
340

345

350

355

360

365

370

375

340

345

350

355

360
Time [s]

355

360
Time [s]

Time [s]

365

370

375

50

40

20
0
0

Gains

Gains

20

40
50
60

80

Feedback Gain
Feedback Gain
Feedforward gain

100

100

Feed back gain

Feed back gain
Feed forward gain

120
0

3.2.0.13

50

100

150

200

250
Time [s]

300

350

400

450

340

345

350

365

370

375

380

Velocity variation rejection and comparison

One of the main characteristics of the MCS control scheme is its capacity to face parameter variations if the
variation of the parameter is slow than the gain adaptation law. Two interesting case are here shown, the rst
shows a 2g deceleration of the air stream and sudden re-acceleration in a steady state, the second case will show
only a 2g deceleration during a transient of a manoeuver.
The rst velocity prole that is used is the shown in g.3.29.

63

MRAC Minimum Controller Synthesis control for flutter

Aistream Velocity
25

U [m/s]

20

15

10

120

122

124

126

128

130
Time [seconds]

132

134

136

138

140

Figure 3.29: Airstream velocity variation

The results of the simulation carried on all the controllers are shown in the following gure

0.2

0.15

0.1

0.05

0.05

0.1

Reference
MCSI
NEMCSI
MCS
LQNEMCSI

0.15

0.2
128.5

129

129.5

130

130.5

131

Figure 3.30: Output for the various controllers under Fig.3.29 airstream parameter variation

The feedback gains are displayed as well for the MCS controller and the LQ-NEMCSI.

64

3.2. MCS control synthesis and simulations

5
4
3

Feedbak gains MCS

2
1
0
1
2
3
4
5
126

128

130

132

134
Time [seconds]

136

138

140

Figure 3.31: Feedback gain for the MCS controller under Fig.3.29 airstream parameter variation

0.3

0.25

Feebback Gains LQNEMCSI

0.2

0.15

0.1

0.05

0.05

128.5

129

129.5

130

130.5
Time [seconds]

131

131.5

132

Figure 3.32: Feedback gain for the LQ-NEMCSI controller under Fig.3.29 airstream parameter variation

65

MRAC Minimum Controller Synthesis control for flutter

x 10
3

NEMCSI KN(t)

127

128

129

130

131
Time [seconds]

132

133

134

Figure 3.33: Kn (t) gain for the LQ-NEMCSI controller under Fig.3.29 airstream parameter variation

As can be seen from Fig.3.31,3.32 and 3.33, while the MCS controller can only compensate the variation
of airstream speed with a Feedback action in the LQ-NEMCSI the variation is compensated main by the
discontinuous action KN (t)sgn(ye ) and the integral action (here not shown).
The seconds parameter variation wave form used is displayed in g.3.34.

Aistream Velocity U

26

24

22

Air stream velocity [m/s]

20

18

16

14

12

10

6
0

50

100
Time [seconds]

Figure 3.34: Airstream velocity variation

The output for the MCS and LQ-NEMCSI controller is shown in Fig.3.35

66

150

3.2. MCS control synthesis and simulations

8
7
6
5

4
3
2
1
0
1

Reference
MCS
LQNEMCSI

126

127

128

129

130
Time [seconds]

131

132

133

134

Figure 3.35: Output for the MCS and LQ-NEMCSI controllers under Fig.3.34 airstream parameter variation

3.2.0.14

Hybrid parameter variation

A simulation that is worth to be displayed is one with a hybrid system. Although this of no physical interest
in this type of system (U or a vary with continuity), it interesting to show how the controller reacts to such
variation. To achieve a hybrid system a sudden variation (a 20 m
s velocity decrease) of U in a null time is
introduced, at t = 126.5[s] as shown in g.3.36

25

U airstream velocity

20

15

10

20

40

60

80

100
Time [s]

120

140

160

180

200

Figure 3.36: Airstream velocity "hybrid" variation

The output of the system is shown in 3.37. The controller does take some time to handle the variation but
the over all behaviour is good, especially considering the tracking error is below 1[degree] approximately. The
gains are also plotted for such variation. A ne tuning allows better performances.

67

MRAC Minimum Controller Synthesis control for flutter

[degrees]

Reference

0
126

128

130

132

134
Time [seconds]

136

138

140

142

144

Figure 3.37: Feedback gains for the chaos recovery simulation

0.1
0.5
0.08
0.06
0
Feedback gains K

0.04

KN(t)

0.02
0
0.02

0.5

0.04
0.06
0.08

1.5

0.1
120

125

130
Time [s]

135

140

120

125

130
135
Time [seconds]

140

145

150

(a) KN (t) gain variation for the closed loop system under

(b) Feedback K gain variation for the closed loop system

going parameter variation in fig.3.36

under going parameter variation in fig.3.36

60

40

Integral gain KI

20

20

40

60

120

125

130

135
Time [seconds]

140

145

(c) Integral gain variation for the closed loop system un-

der going parameter variation in fig.3.36

68

3.2. MCS control synthesis and simulations

3.2.0.15

Chaos recovery

As mentioned in chapter 1 we proposed to show how the MCS strategy - or an extension - can control and
recover from chaos. In Fig. 3.38, the system is left to its chaotic behaviour for 50 seconds an then the MCS
controller is switched on. The system is simulated with no saturations.

80

60

d/dt [degrees/sec]

40

20

20

40

60

80
4

[deg]

10

Figure 3.38: Chaos recovery

The system after the chaotic evolution is to follow the reference in Fig.3.4. After a slow transient, the chaotic
vector eld is removed an tracking is well achieved. A major aspect to consider in the simulation it very high
value for control signal as shown in Fig.3.39 which has to be saturated.

1400

1200

1000

[deg]

800

600

400

200

200
60

80

100
Time [seconds]

120

140

160

Figure 3.39: Control signal for the simulation on chaos recovery

A saturation is thus introduced and the control signal is displayed in Fig. 3.40 as well with the phase plane
in Fig.3.41.

69

MRAC Minimum Controller Synthesis control for flutter

30

20

[degrees]

10

10

20

30

20

40

60

80

100

120

140

160

180

Time [seconds]

Figure 3.40: Control signal with saturation for the simulation on chaos recovery

80

60

40

d/dt

20

20

40

60

80
4

10

Figure 3.41: Chaos recovery

Finally the gains for the simulation with the saturation are also shown in the following gures:

70

12

3.2. MCS control synthesis and simulations

20

Feedback Gains

15

10

20

40

60

80

100

120

140

160

180

140

160

180

Time [seconds]

Figure 3.42: Feedback gains for the chaos recovery simulation

Feedforward Gain

50

100

150

200

250

20

40

60

80

100

120

Time [seconds]

Figure 3.43: Feedback gains for the chaos recovery simulation

71

MRAC Minimum Controller Synthesis control for flutter

72

References
[1] Mario Di Bernardo. Non linear dynamics and control lectures. University Federico II, Naples Italy, 2012.
[2] Bigoni D. e Noselli G. Experimental evidence of utter and divergence instabilities induced by dry friction.
Journal of mechanics and physics of solids, 2011.
[3] Y. C. Fung. An Introduction to the Theory of Aeroelasticity. John Wiley and Sons New York, 1955.
[4] M. Golubitsky. Bifurcation theory. Ohio State University, 2011.
[5] Alberto Isisdori. Nonlinear Control systems. Springer, 1995. from page 137 to 144.
[6] I. D. Landau. Adaptive Control: The Model Reference Approach. Marcel-Dekker, 1979.
[7] Umberto Montanaro Mario di Bernardo, Alessandro di Gaeta and Stefania Santini. Synthesis and experimental validation of the novel lq-nemcsi adaptive strategy on an electronic throttle valve. IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, 2010.
[8] ONeil and W. Strganac. An experimental investigation of nonlinear aeroelastic respons. AIAA Journal of
Aircraft, 1995.
[9] C. Gilliatt ONeil and T. W. Strganac. Investigations of aeroelastic response for a system with continuous
structural nonlinearities. In 37th AIAA Structures, Structural Dynamics and Materials Conference, Salt
Lake City, Utah, 1996.
[10] Nicola Schiavoni Paolo Bolzern, Riccardo Scattolini. Fondamenti di controlli automatici. Mc Graw-Hill,
3rd edition, 2008.
[11] D. P. Stoten and H. Benchoubane. Empirical studies of a mrac algorithm with minimal controller synthesis.
International Journal of Control, 1990.
[12] D. P. Stoten and H. Benchoubane. Robustness of a minimal controller synthesis algorithm. International
Journal of Control, 1990.
[13] S. Strogaz. Nonlinear dynamics and chaos: with applications to physics, biology, chemistry, and engineering.
Perseus Books, 1994.

73

REFERENCES

74

List of Figures
1.1

Lift (L), resistance (R) and pitch moment (P) of a foil in a steady airstream . . . . . . . . . . . .

10

1.2

Typical Cl ,Cm and Cr behaviour . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

10

1.3

Foil subject to elastic recall K . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

11

1.4

Foil subject to elastic recall K in opposite direction after overshoot . . . . . . . . . . . . . . . . .

11

1.5

The tail of the Piper during an LCO . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

12

1.6

The Ziegler Column 2 DoF arm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

13

1.7

Schematic representation of the 2 DoF mechanical arm. vp is the plate velocity . . . . . . . . . .

13

1.8

Experimental data and model prediction for the Ziegler Column . . . . . . . . . . . . . . . . . .

13

1.9

Conguration of the nonlinear 2-D prototypical aeroelastic wing

. . . . . . . . . . . . . . . . . .

14

1.10 Wing cross-section representation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

15

1.11 Wing cross-section schematic representation showing a and mid-chord b . . . . . . . . . . . . . .

15

1.12 Two trailing edge control surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

16

1.13 Simulink block diagram for the systems open loop dynamics . . . . . . . . . . . . . . . . . . . .
1.14 Open loop responses on AoA for the system with = 0.1[rad], h = 0.1[m] and = h = 0 and
U = 15, 17, 20, 19, 40, a = 0.4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.15 Open loop responses on plunge for the system with = 0.1[rad], h = 0.1[m] and = h = 0 and

18

U = 15, 17, 20, 19, 40, a = 0.4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1.16 Open loop responses on AoA for the system with = 0.1[rad], h = 0.1[m] and = h = 0 and
a = 0.1, 0.3, 0.4, 0.6 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.17 Open loop responses on AoA for the system with = 0.1[rad], h = 0.1[m] and = h = 0 and

20

a = 0.1, 0.3, 0.4, 0.6 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

21

2.1

SS graph for AoA partial FBL with = 0.1[rad], h = 0.01[m] and = h = 0 . . . . . . . . . . .

27

2.2

FBL control scheme for AoA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

27

2.3

Phase Plane of the controlled AoA dynamics with the use of a LQR controller . . . . . . . . . . .

28

2.4

Step response of the controlled AoA dynamics with the use of a LQR controller . . . . . . . . . .

28

2.5

Time response and control input for = z1 = x2 and = z2 = x4

. . . . . . . . . . . . . . . . .

29

2.6

SS graph for the plunge after partial FBL on the AoA. It is clear that the variable is indirectly
inuenced . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

30

2.7

Tracking graph and reference signal with null initial conditions . . . . . . . . . . . . . . . . . . .

30

2.8

Plunge position and speed with null initial conditions when the AoA subject to reference signal
as in Fig. 2.7 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

31

Tracking error for give by building FBL control law on dierent parameters (UF BL , aF BL ) from
the systems one (U,a) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

31

2.10 dynamics when subject to non rejected parameter variations . . . . . . . . . . . . . . . . . . .

32

2.11 Null tracking error for subject to parameter variation with a PD controller linear controller . .

32

2.12 Plot of the eigenvalues with respect to velocity U and a. Color pattern towards red indicates
increasing values for the eigenvalues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

33

2.9

75

20

21

LIST OF FIGURES
2.13 Phase plane for a closed loop response for partial FBL on the plunge dynamics, with the following
initial conditions h = 0.1 and = = h = 0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.14 Time response for partial FBL on the plunge dynamics, with the following initial conditions
h = 0.1 and = = h = 0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.15 Time response for partial FBL on the plunge dynamics, with the following initial conditions
h = 0.1 and = = h = 0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.16 Phase plane for and for a closed loop response for partial FBL on the plunge dynamics, with
the following initial conditions h = 0.1 and = = h = 0 . . . . . . . . . . . . . . . . . . . . . .
2.17 Lyapunov function for the plunge Zero dynamics. It is clearly continuous, dierentiable, null in
(0, 0), positive denite and radially unbounded . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.18 Derivative of the Lyapunov function of Fig.2.18. . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.19 Bifurcation diagram with respect to and with a = 0.4 . . . . . . . . . . . . . . . . . . . . . .
2.20 Bifurcation diagram with respect to and with a = 0.5 . . . . . . . . . . . . . . . . . . . . . .
2.21 Bifurcation diagram with respect to and with a = 0.6 . . . . . . . . . . . . . . . . . . . . . .
2.22 Bifurcation diagram with respect to and with a = 0.63 . . . . . . . . . . . . . . . . . . . . . .
2.23 Bifurcation diagram with respect to a and with = 115 . . . . . . . . . . . . . . . . . . . . . . .
2.24 Bifurcation diagram with respect to a and with = 115 . . . . . . . . . . . . . . . . . . . . . . .
3.1
3.2
3.3
3.4
3.5
3.6
3.7
3.8
3.9
3.10
3.11
3.12
3.13
3.14
3.15
3.16
3.17
3.18
3.19
3.20
3.21
3.22
3.23
3.24
3.25
3.26
3.27
3.28
3.29
76

MCS scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
LQMCS scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Reference model step response . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Reference signal period . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Reference signal and output for a MCS controller on the AoA dynamics . . . . . . . . . . . . . .
Control signal for a MCS controller on the AoA dynamics . . . . . . . . . . . . . . . . . . . . . .
Tracking error for a MCS controller on the AoA dynamics . . . . . . . . . . . . . . . . . . . . .
Feed Forward gains for the single MCS controller on the AoA dynamics . . . . . . . . . . . . . .
Feed back gains for the single MCS controller on the AoA dynamics . . . . . . . . . . . . . . . .
Feed back gains for the single MCS controller on the AoA dynamics . . . . . . . . . . . . . . . .
Control signal and relative Feedback gain for a single maneuver . . . . . . . . . . . . . . . . . . .
Reference signal and output for a single MCSI controller on the AoA dynamics . . . . . . . . . .
Tracking error for a single MCSI controller on the AoA dynamics . . . . . . . . . . . . . . . . . .
Control signal for a single MCSI controller on the AoA dynamics . . . . . . . . . . . . . . . . . .
Feed back gain for the single MCSI controller on the AoA dynamics . . . . . . . . . . . . . . . .
Feed Forward gains for the single MCSI controller on the AoA dynamics . . . . . . . . . . . . . .
Integral gain KI for the single MCSI controller on the AoA dynamics . . . . . . . . . . . . . . .
MCS vs MCSI controller on AoA dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Reference signal and output for a single EMCSI controller on the AoA dynamics . . . . . . . . .
EMCSI Feedback gains for AoA dynamics with one control surface . . . . . . . . . . . . . . . . .
EMCSI gains for AoA dynamics with one control surface . . . . . . . . . . . . . . . . . . . . . . .
Control signal for a single EMCSI controller on the AoA dynamics . . . . . . . . . . . . . . . . .
Controller tracking comparison between MCSI and EMCSI controller on the AoA dynamics after
a longer simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Output performance comparison between NEMCSI and EMCSI . . . . . . . . . . . . . . . . . . .
NEMCSI controller gains on the AoA dynamics after a gain assessment, with exception of KN
reported in Fig.3.26 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
NEMCSI controller KN gain on the AoA dynamics after a gain assessment . . . . . . . . . . . .
LQ-NEMCSI controller performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Gain plot for LQ-NEMCSI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Airstream velocity variation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

35
35
36
36
37
38
39
39
40
40
41
41
44
45
47
49
50
50
51
51
52
52
53
53
54
54
55
55
56
57
58
58
59
59
60
60
61
61
62
62
64

LIST OF FIGURES
3.30
3.31
3.32
3.33
3.34
3.35
3.36
3.37
3.38
3.39
3.40
3.41
3.42
3.43

Output for the various controllers under Fig.3.29 airstream parameter variation . . . . . . . . .
Feedback gain for the MCS controller under Fig.3.29 airstream parameter variation . . . . . . .
Feedback gain for the LQ-NEMCSI controller under Fig.3.29 airstream parameter variation . .
Kn (t) gain for the LQ-NEMCSI controller under Fig.3.29 airstream parameter variation . . . .
Airstream velocity variation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Output for the MCS and LQ-NEMCSI controllers under Fig.3.34 airstream parameter variation
Airstream velocity "hybrid" variation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Feedback gains for the chaos recovery simulation . . . . . . . . . . . . . . . . . . . . . . . . . .
Chaos recovery . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Control signal for the simulation on chaos recovery . . . . . . . . . . . . . . . . . . . . . . . . .
Control signal with saturation for the simulation on chaos recovery . . . . . . . . . . . . . . . .
Chaos recovery . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Feedback gains for the chaos recovery simulation . . . . . . . . . . . . . . . . . . . . . . . . . .
Feedback gains for the chaos recovery simulation . . . . . . . . . . . . . . . . . . . . . . . . . .

.
.
.
.
.
.
.
.
.
.
.
.
.
.

64
65
65
66
66
67
67
68
69
69
70
70
71
71

77

List of Tables
1.1
1.2

System SS variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
System SS parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

79

17
17