Sie sind auf Seite 1von 6

Stability Analysis and Backstepping

Controller design for a F-16 Aircraft


Viana. P ∗

Facultad de Minas, Universidad Nacional de Colombia, Medelln
(e-mail: jpvianav@unal.edu.co).

Abstract: This work describes the dynamic non-linear model of the longitudinal flight variables
of an aircraft F-16 fighter. A stability analysis is performed using the Lyapunov theory over its
linearized version. A backstepping controller is designed using a simplified model to force the
aircraft to follow a certain path angle commanded by the pilot, the results were positive within
the range defined for the controller.

Keywords: Non-linear control, Lyapunov controller, flight controller, Lyapunov stability,


Backstepping.

1. INTRODUCTION

The main function of a flight control system is to con-


tribute to the safe an economic operation of the aircraft
by providing tight tracking of the pilot commands and in
many cases reducing the fuel consumption. In the military
industry it is unthinkable to fly an aircraft without a
controller system, mainly because of the high quantity
of decisions the pilot would have to make for a single
maneuver at high speed (Magni et al., 1997).

In the search for competitiveness, several aircraft manu-


facturers are leaving behind the classic control paradigm
and are entering in the modern control era as an effort
to reduce costs and to guarantee robustness before distur- Fig. 1. F-16 Falcon Fighter
bances and modeling uncertainties. Some common control
techniques in the aeronautic industry as non-linear damp- • and finally approximations using backstepping are
ing inversion, fuzzy logic and Lyapunov based techniques presented by (Härkegård, 2001) and (Tran, 2016).
are well documented in (Magni et al., 1997). This work aims to describe the non-linear dynamic model
The F-16 Falcon (see Figure 1) is a supersonic multi role of an F-16 fighter, to perform a stability analysis of the
fighter aircraft originally developed for the US Air-Forces system for a certain flight conditions and to propose a
as an air superiority fighter. Since its first flight in 1978 the Lyapunov based backstepping controller for the trajec-
original model have been through several modifications, tory angle of the aircraft. The development reported by
mostly in computational tools, but the essence remains. (Härkegård, 2001) and (Tran, 2016) will be the basis for
the work here presented, but this work does not deal with
The dynamic parameters of F-16 aircraft have been ob- the dynamic and restrictions of the actuators.
tained experimentally by NASA using wind-tunnel tests ,
the information is available for public in (Nguyen, 1979),
and thats the reason why the F-16 have been so widely This document is organized as follow: in section 2 the
used as test-bed for several developments in flight control. dynamic model and assumptions are presented, 4 gives a
stability analysis based in the Lyapunov theory, section
5 exposes the development of the backstepping controller
Non-linear techniques for flight control applied explicitly and section 6 presents the conclusions of this work.
to the F-16 fighter have been reported in the literature:
• A sliding-mode method is presented by (Seshagiri and 2. AIRCRAFT F-16 DYNAMIC MODEL
Promtun, 2008),
• A non-linear adaptive control is presented by (Sonn- The dynamic model of the aircraft will follow the mathe-
eveldt et al., 2009), matical description given by (Härkegård, 2001) and (Tran,
• Predictive approaches as the Receding Horizon Con- 2016). The model arises from applying the second New-
trol in (Keviczky and Balas, 2006a) ton’s law on a certain reference frame to obtain a set of
force equations and a set of moment equations. From now As in the F-16 the engine is located close to the center of
a coordinate system with origin in the aircraft center of mass, the term ZT P in (1c) can be neglected (Seshagiri
gravity will be used and will be referred as ”Body Axis and Promtun, 2008). The remaining term M/Iy will be
System”. simply referred as u.
The Body-Axis System is decomposed as follow:
The lift force L in 1a can be expressed as:
• xB -axis lies parallel to the longitudinal axis of the
aircraft,
• zB -axis lies in the plane of symmetry of the aircraft 1 2
ρV SCL (α) = L̃CL (α) (2)
and 2 T
• yB -axis is orthogonal to both of them where:
The dynamic of the aircraft can be segregated into the • rho is the air density [kg/m3 ],
longitudinal and the lateral direction. The controller to be • S is the wing area [m2 ] and,
designed is intended to act on the longitudinal variables • CL is known as the lift coefficient and is highly
shown in Figure 2. dependent of the attack angle.
The lift coefficient of F-16 aircraft was measured by NASA
and reported graphically in (Nguyen, 1979); the reported
data was here approximated using the quadratic function
of the form CL (α) = η1 α2 + η2 α, with η1 = −3.29 and
η2 = 5.0(see Figure 3).

Lift coefficient - quadratic approximation


2

1.8

1.6

1.4

Fig. 2. Longitudinal variables (taken from (Härkegård, Measurements


1.2
2001)) Approximation
CL

1
When the speed of the aircraft is considered to be constant
0.8
(it is usually controlled independently), the longitudinal
dynamic model is described by (1). 0.6

0.4
1
γ̇ = (L + FT sin(α) − mgcos(γ)) (1a) 0.2
mVT 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
(rad)
θ̇ = q (1b)
1 Fig. 3. Quadratic approximation - Lift Coefficient
q̇ = (M + FT ZT F ) (1c)
Iy (Nguyen, 1979)
where: The model in (3) will be used as plant.
• VT is the speed of the aircraft [m],
• FT is the force generated by the engine or thrust 1
force [N], γ̇ = (L̃CL + FT sin(α) − mgcos(γ)) (3a)
mVT
• m is the mass of the aircraft [kg],
• L is the force pushing the aircraft upward or lift θ̇ = q (3b)
force [N], q̇ = u (3c)
• g is the gravitational constant [m/s2 ],
• γ is the angle between the speed vector and the where from Figure 2 arises the relation α = θ − γ.
horizon or path angle [rad],
• α is the angle between the xB axis and the speed 3. LINEARIZATION
vector or the attack angle [rad],
• θ is the pitch angle [rad],
• q is the pitching rate [rad/s], The non-linear system (3) will be linearized to the form:
• Iy is the moment of inertia with respect to the yB
axis [kg/m2 ], ∆ẋ = A∆x + B∆u (4)
• M is the input torque generated by the elevators
(actuator) [kg m] and Where A and B are the Jacobian of (3) with respect to
• ZT F is the offset of the engine position with respect the states and the inputs respectively.
to the mass center.
The linearization will be carried out using the flight decreased to 0 again at 6 seconds, the response is shown
conditions shown in Table 1 and taken from (Keviczky in Figure 4.
and Balas, 2006b).
Path angle -
Table 1. Stable state flight conditions 0.5

(rad)
0 Non-linear
Parameter value Linear
FT 9995.15 [N] -0.5
0 2 4 6 8 10
VT 176.48 [m/s] Time(s)
γ 0 [rad] Pitch angle -
0.5

(rad)
The aircraft parameters are taken from (Nguyen, 1979) Non-linear
Linear
and shown in Table 2. 0
0 2 4 6 8 10
Table 2. F-16 aircraft parameters Time(s)
Pitch rate - q
0.1
Parameter value

q (rad/s)
0.05
m 9305 [kg] Non-linear
Linear
S 27.87 [m2 ] 0
0 2 4 6 8 10
Time(s)
System (3 ) is set to zero to obtain the equilibrium point.

1 Fig. 4. Linearized system response


0= (L̃CL (α) + FT sin(α) − mgcos(γ)) (5a)
mVT
0=q (5b) The linearized system follow properly the non-linear sys-
tem during the simulation time. From the physics of the
0=u (5c) model it was expected to be unstable, once the pitch rate
The equilibrium point will be denoted x0 and numerically establishes at a certain value the aircraft will continue
it has been found to be: to rotate in absence of any controller; this issue will be
analyzed in the following section.
0
" #
4. STABILITY ANALYSIS
x0 ≈ 0.0472 (6)
0 The interest now is to study the origin as an equilibrium
The Jacobian of (3) with respect to the states is: point. From (6) it is seen that the origin is not an equilib-
rium point by itself, so system (3 ) will be transformed.
Consider the deviation variable :
a11 a12 a13
" #
δθ = θ − θ0 (10)
A = a21 a22 a23
a31 a32 a33 where θ0 = 0.0471 is the equilibrium value.
1 The obtained transformed system with equilibrium point
a11 = (−L̃(2η1 α + η2 ) − FT cos(α) + mgsin(γ)) at the origin is:
mVT
1
a12 = (L̃(2η1 α + η2 ) + FT cos(α))
mVT 1
a13 = 0 γ̇ = (L̃CL (δθ, θ0 , γ)
mVT (11a)
a21 = 0 +FT sin(δθ + θ0 − γ) − mgcos(γ))
a22 = 0 δ θ̇ = q (11b)
a23 = 1
a31 = 0 q̇ = u (11c)
a32 = 0 And for simplicity (11) is written as:
a33 = 0
(7)
ẋ = Ax + Bu (12)
The Jacobian of (3) with respect to the input is:
Where its matrix A is:
0
" #
−1.14 1.14 0
" #
B= 0 (8)
A= 0 0 1 (13)
1
0 0 0
And evaluating the equilibrium point and the parameters
the linearized system is: The computed eigenvalues of A are:
−1.14 1.14 0 0
" # " #
∆ẋ = 0 0 1 ∆x + 0 ∆u (9) λ1,2 = 0
0 0 0 1 (14)
λ3 = −1.14
In order to test the performance of the linearized system The stability of system (3) will be checked indirectly by
it is subjected to a change in the stable state pitching checking the stability of its linearized version (11). Accord-
input, it is increased from 0 to 0.1 at 5 seconds and then ing with the The following theorem has been extracted
from (Khalil, 1996): 5.1 Controller design

The design of the controller arises from the following


Theorem: The equilibrium point of a system of the form
assumptions:
ẋ = Ax is stable if and only if all eigenvalues of A satisfy
Reλi ≤ 0 and every eigenvalue with Reλ = 0 has an • The aircraft speed VT and thrust force FT will be
associated Jordan block of order one. The equilibrium point controlled independently and can be considered as
x = 0 is globally asymptotically stable if and only if all constant.
eigenvalues of A satisfy Reλi < 0. • The measured lift coefficient CL reported by (Nguyen,
1979) can be approximately fitted using a linear
As not all the eigen values of A have a negative real
part, from the Lyapunov method it is concluded that the function of the form CL = m1 L̃α for a range of angles
system is not asymptotically stable. The analysis can be from 0 to π/6, where L̃ = 12 ρVT2 S.
complemented by finding the associated Jordan matrix J • The term FT sin(α) in (3) can also be approximated
of A: to a linear function of the form y = m2 FT α for angles
"
01 0
#   between 0 to π/6.
J1,2 0
J(A) = 0 0 0 = (15)
0 J3 Lift Coefficient - C
0 0 −1.14 3
L

The dimension of the Jordan block J1,2 associated to the 2.5


eigenvalue λ1,2 is two, so the conditions of the stability
theorem are not met and can be concluded that the origin 2
of (11) is not stable, and therefore the system (3) neither.
CL 1.5
The effect of having eigenvalues with zero real part and
multiplicity greater than one can be appreciated when 1
solving the differential equation system.
0.5

x(t) = x(t)uf + x(t)f


0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
1
 
(rad)
eλ3 t 1 − eλ3 t t + λ3 eλ3 t + (16)
x(t)uf = 0
 λ3  x(0)
1 t  Fig. 5. Linear approximation - Coefficient (Nguyen, 1979)
0 0 1
If the aim is to control the path angle it would be
where x(t)uf is the unforced response and x(t)f is the convenient to eliminate from (3) the variables whose
forced response. As λ3 has a negative real part, its effects dynamic is not explicitly given.The model for design
will decay with time, but due to the presence of eigenvalues purposes is complete.
con real part zero and multiplicity greater than one the
unforced response has terms that grow with time and make 1
the total response unstable. γ̇ = (m1 L̃(θ − γ) + m2 FT (θ − γ) − mgcos(γ)) (18a)
mVT
θ̇ = q (18b)
5. BACKSTEPPING CONTROLLER q̇ = u (18c)
M
The system (3) is said to be a third order strict-feedback where u = Iy stands for the effect of the pitch torques.
SISO system of the form:
For simplicity purpose system (18) is re-written using
matrices:
ẋ1 = f (x1 , x2 )
ẋ2 = f (x1 , x2 , x3 ) (17)   " #
ẋ2 = f (x1 , x2 , x3 , u) γ̇ fγ
 θ̇  = fθ (19)
q̇ fq
The objective is to find “virtual control laws” using Lya- The following procedure is based on (Tran, 2016).
punov functions in order to lead the system to stability
from the first to the last equation and this process is called Step 1 Let the variable γ̃ represent the deviation from
backstepping. The theory of the backstepping method can the desired flight path:
be found in (Isidori, 1995).
γ̃ = γ − γref (20)
A Backstepping controller which aims to keep the path
angle γ as close as possible to a desired value γref is And replacing into 3 and consider the resulting subsystem
developed below: has θ as control law:
1 From (27) it is possible to calculate α̇2 :
γ̃˙ = (m1 L̃(θ − (γ̃ + γref ))
mVT (21)
α̇2 = −c2 (q − α̇1 ) + α̈1 (30)
+m2 FT (θ − (γ̃ + γref )) − mgcos(γ̃ + γref ))
Where from (28) it is possible to obtain α̈1 :
1 2
Suppose there exists a Lyapunov function V1 (γ) = 2 γ̃
  
∂ α̇1 ∂ α̇1 fγ
whose time derivate is: α̈1 = (31)
∂γ ∂θ fθ

V̇1 (γ̃) = γ̃ γ̃˙ (22) The control law u can be obtained by means of a Lyapunov
function V3 (γ̃, θ̃, q̃) = V2 (γ̃, θ̃) + 21 q̃ 2 and its time derivate:
In order to ensure stability an alternative is setting V̇1 =
−c1 γ̃ 2 , i.e. setting γ̃˙ = −c1 γ̃. By doing so it is possible to V̇3 (γ̃, θ̃, q̃) = V̇2 (γ̃, θ̃) + q̃(u − α̇2 ) (32)
solve for θ to obtain the control law θref that stabilizes The stability conditions in the sense of Lyapunov are met
the subsystem. if the control law u is selected to be:
−c3 q̃ = u − c2
θref = α1 (c1 , γ, γref ) (33)
u = α3 (c1 , c2 , c3 , γ, θ, q, γref ) = −c3 (q − α2 ) + α̇2
 
θref = β1 −c1 mVT γ̃ + (L̃m1 + FT m2 )γ + mgcos(γ) The complete system have been transformed into the state
decoupled system:
1 γ̃˙ = −c1 γ̃ (34a)
β1 =
L̃m1 + FT m2 ˙
(23) θ̃ = −c2 θ̃ (34b)
q̃˙ = −c3 q̃ (34c)
Step 2 Let θ̃ be the deviation of θ from the desired value
θref : And for c1 , c2 , c3 > 0 the system is asymptotically stable
given that it has been granted that the Lyapunov functions
fulfills the conditions of Lyapunov stability theorem:
θ̃ = θ − θref = θ − α1 (24)
Whose time derivate is: V1 (γ̃) > 0
˙ V̇1 (γ̃) < 0
θ̃ = q − α̇1 (25)
1 2
A Lyapunov function V2 (γ̃, θ̃) = V1 (γ̃) + 2 θ̃ whose time V2 (γ̃, θ̃) > 0
derivate is: V̇2 (γ̃, θ̃) < 0 (35)

V˙2 (γ̃, θ̃) = V˙1 (γ̃) + θ̃(q − α̇1 ) (26) V3 (γ̃, θ̃, q̃) > 0
V̇3 (γ̃, θ̃, q̃) < 0
Again, the goal is finding a virtual control law that fulfills
the drive the subsystem to meet the stability requirements
in the sense of Lyapunov:
5.2 Simulation

−c2 θ̃ = q − α̇1 The architecture designed in the previous section is imple-


(27) mented in Simulink - Matlab using the flight conditions
qref = α2 (c1 , c2 , γ, θ, γref ) = −c2 (θ − α1 ) + α̇1 in section 3. Several alternatives for tunning the controller
are given by (Tran, 2016), but for the illustrative purpose
where: of this work a set of gains will be taken directly from (Tran,
∂α1 2016) without deeper insight, they are shown in Table 3.
α̇1 = fγ
∂γ Table 3. Controller parameters
∂α1 h i
Parameter value
= β1 −c1 mVT + (L̃m1 + FT m2 ) − mgsin(γ)
∂γ c1 1.72
(28) c2 1.73
c3 1.72
Step 3 At this stage the goal is to find a control law
u that drives the overall system to the desired path angle During the simulation the aircraft is expected to:
γref . Consider the following change of coordinates and find • Rise its path angle form 0 rad to 0.5 rad without
its time derivate: modifying the speed vector.
• Decrease the path angle from 0.5 rad to 0.3 rad
q̃ = q − α2 without modifying the speed vector.
(29) • Return to the initial condition without modifying the
q̃˙ = u − α̇2 speed vector.
No saturation in the actuators is checked. The path angle 2
Input signal - q
response is shown in Figure 6. u

Path angle - 1.5


0.6

ref
1
0.5

)
0.5

2
u (rad/s
0.4

0
(rad)

0.3
-0.5

0.2
-1

0.1 -1.5
0 5 10 15 20 25 30 35 40
Time (s)

0
0 5 10 15 20 25 30 35 40 Fig. 8. Input signal
Time (s)
pilot, the results were positive for the presented conditions
Fig. 6. Path angle aircraft response and assumptions.

The virtual control laws and their actual response is shown REFERENCES
in Figure 7.
Härkegård, O. (2001). Flight Control Design Using Back-
Pitch angle -
1.5 stepping. Ph.D. thesis, Linkping UniversityLinkping
University, Automatic Control, The Institute of Tech-
1 ref
nology.
(rad)

0.5
Isidori, A. (1995). Nonlinear Control Systems. Communi-
cations and Control Engineering. Springer, 3rd edition.
0 Keviczky, T. and Balas, G.J. (2006a). Receding horizon
control of an f-16 aircraft: A comparative study. Control
-0.5
0 5 10 15 20 25 30 35 40 Engineering Practice, 14(9), 1023–1033.
Time (s) Keviczky, T. and Balas, G.J. (2006b). Receding
Pitch rate - q horizon control of an F-16 aircraft: A comparative
2
q study. Control Engineering Practice, 14(9), 1023–
q
1 ref 1033. doi:10.1016/j.conengprac.2005.06.003. URL
q (rad/s)

http://www.sciencedirect.com.ezproxy.unal.edu.co/scien
0
Khalil, H.K. (1996). Noninear Systems. Prentice-Hall,
-1 New Jersey.
Magni, J.F., Bennani, S., and Terlouw, J. (1997). Robust
-2 flight control: a design challenge, volume 110. Springer.
0 5 10 15 20 25 30 35 40
Time (s) Nguyen, L. (1979). Simulator study of stall/post-stall
characteristics of a fighter airplane with relaxed longi-
Fig. 7. Pitch angle and pitch rate response tudinal static stability.
Seshagiri, S. and Promtun, E. (2008). Sliding
The input signal is presented in Figure 8. As afore men- mode control of F-16 longitudinal dynamics.
tioned no saturation is checked. American Control Conference, 2008. URL
http://ieeexplore.ieee.org/abstract/document/4586748/.
6. CONCLUSIONS Sonneveldt, L., Van Oort, E., Chu, Q., and Mulder, J.
(2009). Nonlinear adaptive trajectory control applied
A dynamic model of the longitudinal variables of an to an f-16 model. Journal of Guidance, control, and
aircraft F-16 has been presented and simplified under the Dynamics, 32(1), 25–39.
assumptions of constant speed and controlled engine force. Tran, T. (2016). Nonlinear Flight Control
Design Using Backstepping Methodology.
The system has been linearized for a certain flight con- Ph.D. thesis, Old Doninion University. URL
ditions and the equilibrium point has been found. The http://digitalcommons.odu.edu/mae etds/16.
equivalence between the linear and the non-linear system
has been exploited to check the stability of the aircraft. It
has been found to be unstable.
A backstepping controller has been proposed to make the
aircraft follow a certain angle path commanded by the

Das könnte Ihnen auch gefallen