Nonlinear, Adaptive Control System For Payload Extraction Operations

ISSN 2176-5480
22nd International Congress of Mechanical Engineering (COBEM 2013)

November 3-7, 2013, Ribeirão Preto, SP, Brazil
Copyright
c 2013 by ABCM
NONLINEAR, ADAPTIVE CONTROL SYSTEM FOR PAYLOAD

EXTRACTION OPERATIONS.
Gustavo Oliveira Violato
Pedro Paglione
Instituto Tecnológico de Aeronáutica - ITA Praça Marechal Eduardo Gomes, 50 - Vila das Acácias CEP 12.228-900 - São José dos
Campos - SP - Brasil
[email protected]; [email protected]
Abstract. The development of a nonlinear, adaptive control system for payload extraction operations is presented. Load
extraction in cargo aircraft during flight is considered as a dangerous maneuver since it normally affects the longitudinal
stability of the aircraft, commonly rendering it unstable. The online adpatation control strategy seems an adequate
approach for solving the problem, since it can deal with the drift in the plant parameters caused by the movement of the
load inside the aircraft. In this work, the effects of a continously varying C.G. position on the longitudinal flight dynamics
are modeled in detail. A controller is then proposed based on the derived equations of motion, applying the technique of
dynamic inversion coupled with a model reference adaptive controller to deal with the varying parameters.The dynamics
system considered for the control problem consists of the modeled aircraft dynamics augmented by the unkown parameters
- whose dynamics are controlled by the chosen adaptation laws. The demonstration of stability for the complete system
is done via Lyapunov’s stability theorem for nonlinear dynamic systems. A suitable Lyapunov Function Candidate is
proposed. Simulation results are presented and discussed based on the theory and the advantages and drawbacks for the
application of such a control law on real aircraft control systems are listed and critically analysed.
Keywords: LAPES, Payload Extraction Systems, Control Design, Adaptive Control.
1. INTRODUCTION
Several kind of payload extraction manouvers are performed as tatical delivery methods in places where there is no
runway for an aircraft to land or the terrain is not adequate for aircraft operations, when precise location of the delivery is
important.
This kind of manouvers can be done with the aid of a LAPES (low altitude payload extraction system) which involves
several systems inside the aircraft - as the parachutes for pulling the load, the pallets and rail system inside the fuselage
and of course some stability augmentation system to help the pilot acomplish this normally dificult task.
Figure 1: A C-130 performing a Low-Altitude Payload Extraction manouver. (U.S. Air Force).
Load extractions are a critical and dangerous manouver, not only they are done in low altitudes and relatively low
velocity, but also, as it shall be shown in section 4., as the load moves inside the aircraft towards its rear, the aircraft’s
center of gravity (CG) gets dislocated, severely affecting the aircraft’s dynamics, sometimes even unstabilizing it. This
4512
ISSN 2176-5480
G. Violato and P. Pagilone
maneuver was already extensively studied from the aircraft control point of view during the decade of the 1960 by works
like Rutan and Stroup (1967).
Since one of the promissing applications of adaptive control is to be able to maintain a controlled plant following
a desired trajectory even in the presence of unpredictable parameter change (as discussed in Astrom and Wittenmark
(2008)), it seems apropriate to study the development of a LAPES with this control technique.
2. DYNAMICS
2.1 LONGITUDINAL AIRCRAFT DYNAMICS
Following the works of Etkin (1972); Roskam (2001); Duke et al. (1988) one can write the nonlinear equations for the
an aircraft’s longitudinal dynamics.
Considering the state variable given by X = [θ q α V H]T and the control variable is given by U = [dp
T
dπ ] the diferential equations defining the longitudinal dynamics is given by:
˙ q
M + zF/CG F cos(αf ) − Iyy
q̇ = (1a)
Iyy
1
V̇ = −D + F cos(αf − α) − g sin(θ − α) (1b)
m
1
α̇ = q − L − m g cos(θ − α) + F sin(αf − α) (1c)
mV
θ̇ = q (1d)
ẋ = V cos(θ − α) (1e)
Ḣ = V sin(θ − α) (1f)
Where the aerodynamic forces and moment are given by:
L = 12 ρV 2 CL D = 12 ρV 2 CD M = 12 ρV 2 c̄CM
The aerodynamic coeficients (CL , CM ), can be reasonably well aproximated to depend linearly on the state variables.
The drag coefficient CD is normally aproximated as a parabolic function of the lift coefficient:
qc̄
CL =CL0 + CLα α + CLq̂ + CLdp dp
2Vref
CD =CD0 + k1 CL + kCL2
α̇c̄ qc̄
CM =CM0 + CMα α + CMα̇ˆ + CMq̂ + CMdp dp
2Vref 2Vref
All the boxed terms is the above equations are dependent on the aircraft’s C.G. location as varied by the displacement
of the cargo load. In the next section the origin and form of this variations is seen in detail.
2.2 EFFECTS OF INTERNAL MASS DISPLACEMENT ON AIRCRAFT DYNAMICS
The dynamics of the aircraft are affected by the displacement of an internal load as it changes the center of gravity
location from its original position. These corrections are presented below.
The moment of inertia of the airplane carrying a load can be expressed as the sum of both the airplane and load inertias
with respect to the current C.G. location:
T otal/Xcg A/Xcg C/Xcg
Iyy = Iyy + Iyy (2)
Each of the terms on the right-hand side of equation 2 can be further expressed as the sum of the moment of inertia of
each component with respect to its own C.G. with the product of its mass to the square of the distance from its C.G. to
the C.G. of the complete system. Defining Xcg0 as the center of mass of the unloaded aircraft and XcgC as the position
of the load’s center of mass we can write:
T otal/Xcg A/Xcg0
Iyy = Iyy + (Xcg − Xcg0 )2 mA + Iyy
C/XcgC
+ (Xcg − XcgC )2 mC (3)
4513
ISSN 2176-5480

C/X
The term Iyy cgC in equation 3 is considered negligible with relation to the others. The displacement of the center of
gravity of the loaded airplane with respect to its original location (Xcg − Xcg0 ) will appear in other expressions as well,
which motivates the definition:
∆Xcg , Xcg − Xcg0 (4)
The current position of center of gravity of the aircraft at a given time is expressed as:
Xcg0 mA + XcgC mC
Xcg = (5)
mA + mC
Defining the aircraft load-mass ratio as µA , mA /(mA + mC ), the relation (Xcg − XcgC ) can be expressed as:
−µA −µA
(Xcg − XcgC ) = (Xcg − Xcg0 ) = ∆Xcg (6)
(1 − µA ) (1 − µA )
Substituting equations 4 and 6 in equation 3 we find:
mA 2
Iyy = Iyy0 + ∆Xcg (7)
1 − µA
A/X
Where the inertia of the empty plane Iyy cg0 was represented as Iyy0 . The derivative of the inertia with respect to time
is obtained directly from equation 7 and is given by:
˙ = 2mA ˙ cg
Iyy ∆Xcg ∆X (8)
1 − µA
Figure 2: Geometry for finding airplane lift and pitching moment derivatives. (Following Roskam (2001))
Variation of CMα Taking the definitions of figure 2 as a base to calculate the balance of longitudinal forces and mo-
ments acting on the aircraft, one arrives at equation 9 for the dependence of the pitching moment with the angle of attack
Roskam (2001); Etkin (1972).
Sh d
CMα |Xcg = CLα wf (X̄cg − X̄acwf ) − CLα h ηh (X̄ach − X̄cg )(1 − ) (9)
S dα
In equation (9), the subscript wf means “the wing-fuselage system” and the subscript h means “the horizontal tail”.
The term ηh is the ratio of the dynamic pressure acting on the horizontal tail with the dynamic pressure acting on the wing
4514
ISSN 2176-5480
(which can be greater than 1 if for example the engines are blowing air right at the tail). Notice also that all distances are
normalized by a reference distance - in this case the mean aerodynamic chord - since equation (9) is dimensionless.
The pitching moment calculated in 9 is with respect to the aircraft’s center of gravity. There is a point of the aircraft
for which the pitching moment does not vary with angle of attack. This point, called the neutral point, can be calculated
from the definition by imposing that CMα |N.P. = 0. So, re-writting 9 for the neutral point one gets:
CLα wf X̄acwf + CLα h ηh SSh (1 − d

dα )X̄ach
X̄N = (10)
CLα wf + CLα h ηh SSh (1 − d
dα )
Defining:
Sh d
CLα 0h , CLα h ηh (1 − ) (11)
S dα
And substituting 10 and 11 in 9 one finds:
CMα |Xcg = (CLα wf + CLα 0h )(X̄cg − X̄N ) (12)

| {z }
CLα
Finally, substituting X̄cg in 12 for X̄cg0 + ∆X̄cg we find the formula for the correction of the derivative of CMα with
C.G. displacement:
CMα |Xcg = CMα |Xcg0 + CLα ∆X̄cg (13)
Variation of CMdp Referring again to figure 2 one expresses the variation of pitching moment with elevator position as
Roskam (2001). After substitution the same substitution for X̄cg done before:
Sh
CMdp |Xcg = − CLα h ηh τe (X̄ach − X̄cg ) = CMdp |Xcg0 + CLdp ∆X̄cg (14)
| {z S}
CLd
p
Variation of CLq̂ and CMq̂ When the aeroplane undergoes a sudden pitch variation, the angle of attack of the horizontal
tail is modified by Roskam (2001); Etkin (1972):
q(Xach − Xcg )
∆α = (15)
V0
This corresponds to an increase in lift:
1 2 q(Xach − Xcg )
∆L = ρV Sh ηh CLα h (16)
2 0 V0
Dividing 16 by 21 ρV02 S one finds:

Sh qc̄
∆CL = 2CLα h ηh (X̄ach − X̄cg ) (17)
| S {z } | 2V 0
{z }
CLq̂ q̂
Thus the variation of CLq̂ with ∆Xcg is given by:
CLq̂ |Xcg0
CLq̂ |Xcg = CLq̂ |Xcg0 − ∆X̄cg (18)
(X̄ach − X̄cg0 )
As for the derivative CMq̂ , it is simply the negative of the product of CLq̂ with (X̄ach − X̄cg ):
Sh
CMq̂ |Xcg = −2.2CLα h ηh (X̄ach − X̄cg )2 (19)
S
4515
ISSN 2176-5480

Where a factor of 10% was added to take into account the effect of the wing on the pitching moment. This is called
the fudge-factor Roskam (2001). Using 18 and 19 one can solve for the horizontal tail distance from the center of mass:
CMq̂ |Xcg0
(X̄ach − X̄cg0 ) = (20)
−1.1CLq̂ |Xcg0
Substituting 20 in equations 18 and 19 one finds:
1.1(CLq̂ |Xcg0 )2
CLq̂ |Xcg = CLq̂ |Xcg0 + ∆X̄cg (21)
CMq̂ |Xcg0
(1.1CLq̂ |Xcg0 )2
CMq̂ |Xcg = CMq̂ |Xcg0 + 2.2CLq̂ |Xcg0 ∆X̄cg + (∆X̄cg )2 (22)
CMq̂ |Xcg0
Variation of CMα̇ˆ As in the case of the CMq̂ derivative, CMα̇ˆ is the negative product of CLα̇ˆ with the distance from
the horizontal tail with the C.G..
This effect is caused manly by the variation in intensity of the vortices shed by the wing with instantaneous angle of
attack. The vortices shed by the wing at a time t will reach the horizontal tail at a time t + ∆t, where ∆t is the time
the flow takes to get from the wing to the horizontal tail. This delay is a funcion of the aicraft’s speed and geometric
properties, as the distance from the aerodynamic centers of the wing and horizontal tail. There is a common practical
approximation in the literature Roskam (2001); Etkin (1972) where this distance is approximated by the distance from the
C.G. to the horizontal tail. While this is a valid approximation for most flight conditions, this is not true in our case.
One can find in the cited references a more in depth description of the calculations for the derivatives CLα̇ˆ and CMα̇ˆ .
We only stress here that CLα̇ˆ does not depend on the aircraft’s center of mass on a first approximation. It only depends
on the geometric properties of the aeroplane.
The dependence of CMα̇ˆ with the C.G. position is thus given by:
CMα̇ˆ |Xcg = −CLα̇ˆ (X̄ach − X̄cg ) = CMα̇ˆ |Xcg0 + CLα̇ˆ ∆X̄cg (23)
2.3 PAYLOD EXTRACTION DYNAMICS
It is necessary to establish what will be the dynamics of the aircraft’s center of gravity position during the time the
load is being ejected from the fuselage as in figure 3.
Figure 3: Schematics for load positions during extraction
Normally load extraction manouvers are acomplished with the use of parachutes that use the air resistance to pull the
payload out of the airplane. Consider the limit case when there is no parachute and the load is free to slide down the rails
only by action of gravity.
As presented in figure 3, for this study, the load’s center of gravity position (XcgC ) is considered to coincide with the
empty aircraft center of gravity location (Xcg0 ) just before the ejection command is issued.
4516
ISSN 2176-5480
Then, the load starts sliding down the rails by action of gravity. Notice that the rails (dotted line in 3) have an
inclination of σ with respect to the longitudinal axis of the airplane (the dashed line). The dynamics are then described by
the classic problem of a mass sliding down an inclined plane with no friction.
Now, back to the dynamic modeling of the sliding load. While inside the aircraft, the load will accelerate according to
24:
ẌcgC = g sin(σ + θ), XcgC < XcgC max (24)
where σ and θ in equation 24 are defined as in figure 3. Note that σ is fixed.

Since the equations are being developed as a funcion of the total CG displacement (∆Xcg ) from its original position
(Xcg0 ), one can use equations 24, 5 and 6 to find the dynamics for ∆Xcg . Substituting 6 for Xcg in 5 one finds:
mA −µA
(Xcg0 − XcgC ) = ∆Xcg (25)
mA + mC (1 − µA )
| {z }
µA
Taking the second time-derivative of 25 and replacing 24 for ẌcgC we have:
∆Ẍcg = (1 − µa )g sin(σ + θ), ∆XcgC < ∆XcgC max (26)
We have thus all the required dynamics necessary to simulate the load extraction from the aircraft.
3. NONLINEAR-ADAPTIVE CONTROLLER DEVELOPMENT
A nonlinear adaptive control law for regulation of the pitch angle using the elevator as the only control input will be
developed. It is further assumed the aircraft has the following sensors:
• A rate gyroscope for measuring the pitch angular rate q

• An integrating gyroscope for mearing the pitch angle θ
• Air sensors (pitot, static intake, thermometer) for measuring the dynamic pressure q̄ and estimating the aircraft’s
velocity V
• An acelerometer at the aircraft’s center of gravity and aligned with the aeroplane Z body axes.
The derivation of the controller follows Lyapunov’s second method as in Singh and Steinberg (1996); Steinberg and
Page (1998); Astrom and Wittenmark (2008). Consider the MRAC regulator architecture given by the diagram 4.
Figure 4: Block diagram for a model-reference adaptive controller (MRAC) for pitch control during load extraction.
The objective is to find a control law δp (q, θ, θref , Â, b̂) dependent on the estimated parameters and adaptation schemes
˙ ˙
Â and b̂ that will stabilize the trajectory in the reference model’s trajectory and adapt the parameters to their real values.
This can be acomplished by defining a Lyapunov candidate function dependent on the trajectory error and the param-
eter’s estimation errors and defining the control and adaptation laws so that its derivative will be negative semi-definite.
First, though, the underlying control problem (as described in Astrom and Wittenmark (2008)) will be solved using
feedback linearization on the fast dynamics. The obtained ideal control will be after compared with the resulting control
dependent on the parameter’s estimates.
4517
ISSN 2176-5480

Feedback Linearization Law: Centainty Equivalence From equations 1d and 1a one finds for the aircrafts fast longi-
tudinal dynamics:
θ̇ = q
˙ q
M + zF/CG F cos(αf ) − Iyy
θ̈ = q̇ = (27)
Iyy
Expanding the aerodynamic moment in its components the second equation in 27 can be written as:

q̄Sc̄ c̄ c̄
θ̈ = CM0 + CMα α + CMα̇ˆ α̇ + CMq̂ q + CMdp dp
Iyy 2V0 2V0
zF/CG F cos αf I˙yy q
+ − (28)
Iyy Iyy
Now, as seen in section 2., the aerodynamic coefficients in (28) will vary with C.G. position. We thus separate the state
variables in one vector and the coefficients in another to get:
θ̈ = A0 + φTA1 A1 + φTb bδp (29)
Where:
zF/CG F cos αf I˙yy q
A0 = −
Iyy Iyy
T
q̄0 Sc̄ c̄ c̄
A1 = CM0 CMα CMα̇ˆ CMq̂
Iyy 2V0 2V0
q̄0 Sc̄
b= CMdp
Iyy
q̄
φA1 = [1 α α̇ q]T
q̄0
q̄
φb =
q̄0
1
q̄0 = ρ0 V02 (30)
2
In equation 3., ρ0 means the air density at sea level and V0 the airspeed for which the aerodynamic coefficients where
estimated.
Let us assume that it is desired to have a dynamics for the pitch angle described by:
θ̈m = −k1 θ̇m − k2 (θm − θref ) (31)
Where k1 and k2 make a stable dynamics the the reference model system.
With perfect knowlegde of the system’s parameters and a perfect state measurement one finds by dynamic inversion
that the control:
1
dp = T −A0 − φTA A − k1 θ̇m − k2 (θm − θref ) (32)
φb b
will make the fast dynamics follow the desired model. Since neither the parameters nor the state are known precisely the
control law given by 32 has to be modified.
Adaptive Feedback Linearization Note first that one can group A0 and A1 by augumenting the regression vector φA1
to get:
θ̈ =φTA A + φTb bdp (33)
T
A =[A0 A1 ] (34)
T
φA =[1 φA1 ] (35)
Now consider we only have estimates of the vectors A and b given by Â and b̂ and that the errors in such estimates are
given by:
Ã =A − Â
b̃ =b − b̂ (36)
4518
ISSN 2176-5480
Now let us introduce the generalized error s given by:

˙ + K(θ − θm ) = q − qm + K(θ − θm )
s = θ̇ − θm (37)
Following works such as Singh and Steinberg (1996); Astrom and Wittenmark (2008), this work proposes the following
Lyapunov candidate function V :
1 2 1 T 1
V = s + Ã P Ã + b̃2
2 2 γ
we find that:
dV 2 ˙
= sṡ + ÃT P Ã˙ + b̃b̃ (38)
dt γ
A hypothesis is need to continue the developement of equation 38. From the definition of the parameters errors in equation
36 it follows that their time-derivatives will be a sum of the actual parameters time variation with the adapted parameters
time variation. The hypothesis assumed is that the true parameters time variation is small enough so that we can consider
the following approximation:
˙
Ã˙ = −Â
˙ ˙
b̃ = −b̂ (39)
This is the same as considering that the parameters are “slow varying” with respect to both the aircraft dynamic modes
and most importantly the controller adaptation dynamics.
Now, the controller being developed is actually able to track a pitch setpoint, as it will follow any desired model
dynamics. For simplicity, let us consider the regulation problem, where qm = 0 and θm = θref . In those conditions,
derivating equation 37 in time gives:
ṡ = q̇ + Kq = φTA A + φTb bdp + Kq (40)
Using 40, 36 and 39 in 38 one finds:
dV
˙ 2 ˙
= s φTA Â + φb b̂dp + Kq + φTA Ã + φb b̃dp − ÃT P Â − b̃b̂ (41)
dt γ
Choosing the control law:
1
dp = −Kq − φTA Â − c1 s (42)
b̂φb
Where: c1 > 0
And substituting in equation 41 we find that:
dV
˙ 2 ˙
= −c1 s2 + s φTA Ã + φb b̃dp − ÃT P Â − b̃b̂ (43)
dt γ
Now we choose the parameter’s adaptations to cancel the unknown terms in equation :
˙
Â = sP −1 φA (44)
˙ γ
b̂ = sφb dp (45)
2
and the derivative of the Lyapunov function becomes:
dV
= −c1 s2 (46)
dt
This function is negative as long as the measured trajectory is not equal the refence trajectory. In the case of regulating the
pitch angle, this will happen as long as the measured pitch angle is not equal to the reference signal or if there is any pitch
rate measured by the rate gyro. With the control law in 42 and adaptations 44 and 45 the traking error will thus always go
to zero.
4519
ISSN 2176-5480

Controller implementation Comparing equations 32 and 42, together with de definition of s in 37, it is found that the
adaptive control law corresponds to the ideal control law, only by exchanging the real parameter’s by the estimated ones.
Notice, however, that according to the definition of φA1 in equations 3., it would be necessary to measure α and α̇
to apply the control law deduced in 42. As those measures are not normally available in standard aircraft, they shall be
replaced by the considered measured variables described in the beggining of this section.
Keeping in mind the regulation problem of keeping the pitch for leveled flight, one finds that for such flight conditions:
α≈θ (47)
Moreover, considering ouput from the accelerometer placed in the aircraft’s CG and aligned with its vertical axis (see
Duke et al. (1988)), and the expression for α̇ as in 1c one finds that for small angular rates q it is reasonable to assume:
g
α̇ ≈ q + (1 − nz ) (48)
V
Where nz is the normalized measure of the accelerometer. Finally, substituing equations 47 and 48 for the expression of
φA one finds the implementation of the control law and parameter adaptations as:
1
dp = −Kq − φTAm Â − c1 s (49)
b̂φb
˙
Â = sP −1 φAm (50)
˙ 2
b̂ = sφb dp (51)
γ

q̄ q̄ q̄ g q̄
Where: φAm = 1 θ q + (1 − nz ) q
q̄0 q̄0 q̄0 V q̄0
K, c1 , P, and γ are parameters chosen by the designer
Equations 49 to 51, together with definitions and 37 form the adaptive control laws for pitch tracking. Which shall be
used for pitch regulation with the choice of qm = 0 and θm = θref , where θref will the the initial trimmed pitch angle in
the following simulations.
4. SIMULATION
The airplane chosen for this study was the Lockheed Martin C-5A Galaxy. A scaled two-sided view of the aircraft and
the parameters for its dynamic model are given in figure ??.
From the dimensions described the schematics, it is possible to infer the length of the path described by the load inside
the aircraft. The relevant parameters for this case study are presented in table 1.
Mass of unloaded aircraft mA 248416kg

Mass of the load mC 15000kg
Length of the rails XcgC max − Xcg0 26.2m
Maximum C.G. displacement ∆Xcg max 1.49m
Initial C.G. position Xcg0 0.30c̄
C.G. position at extraction Xcgmax 0.46c̄
Inclination of cargo ramp σ 5◦
Table 1: Parameters of load extraction for the C-5A Galaxy.
The aircraft’s parameters used in this work belongs to the Lockheed Martin’s C-5A Galaxy in “Power Approach
Configuration” as presented in Heffley and Wayne (1972).
The simulations consisted in implementing the dynamic equations for longitudinal flight with varying center of gravity
developed in section 2.. The C.G. dynamics was implemented as described in subsection 2.3. First the trim commands and
states were found for a nominal flight condition as described in table 2.
During the simulation of the aircraft’s flight, equation 26 can be used together with initial conditions ∆Xcg (T E − ) =
∆Xcg0 and ∆Ẋcg (T E − ) = 0 to calculate the current position of the CG at each time step after the extraction comand
instant (T E).
Once the load is extracted (XcgC = XcgC max ), the simulation must restart from its current state but considering the
coefficients, mass and inertia for the unloaded aircraft. This “restart” is accomplished in the software Matlab
R
by the use
of the “Events” option of the ode45 option set Inc. (2000-2012) in the following manner:
4520
ISSN 2176-5480
Figure 5: Top and side view of the C-5A. (Heffley and Wayne (1972))
Leveled Flight Conditions Trimmed State and Commands

H 500 m α=θ 0.68◦
V 75 m/s dp 0.06◦
∆Xcg 0m Throttle 39.8%
Table 2: Nominal flight conditions before load extraction
• The simulation starts with the load’s C.G. at the same position as the aircraft’s C.G.
• At time T E, an extraction command is issued and the load starts sliding down the rails following the dynamics 24
• At each integration step, an event function is called to check the load’s C.G. position and compare it to its maximum
limit XcgC max
• When this limit is reached, the integration is stoped. The current time and state are saved and a “load dropped flag”
is set to True.
• If the current time is less then the final integration time, which was previously chosen, the integration is then
restarted with initial conditions equal to the saved simulation state just before the drop.
5. RESULTS
Two cases were studied. One with fixed controls at the trimmed positions for later reference and one with the adaptive
controller in action. Three gain sets were applied on the second case. Simulations were run starting from nominal flight
conditions and “activating” the load dynamics at “T = 5s” (for the fixed controls) or “T = 7s” (with the adaptation
scheme).
Fixed commands Figure 6 shows the simulation results on the states for the load extraction with fixed controls. The
red dot in the graphics shows the instant when the load was extracted. The influence of the CG variation is seen clearly
by the large fluctuations of angle of attack during and after the load movement.
The natural, fixed control variation of the parameters which will be adapted is shown in figure 7. Notice the sharp
change in the parameters values at the moment of extraction.
4521
ISSN 2176-5480

800 80 15
600 60 10
V (m/s)
H (m)
α (°)
400 40 5
200 20 0
0 0 −5
0 10 20 30 0 10 20 30 0 10 20 30
T (s) T (s) T (s)
2 40 20
30 15
1.5
20 10
(m)
q (°/s)
θ()
10
°
5
cg
∆X
0.5 0
0
0 −10
−5
−20
−0.5 −10
0 10 20 30 0 10 20 30 0 10 20 30
T (s) T (s) T (s)
Figure 6: Load extraction simulation with fixed commands. Evolution of the states.
−3
x 10
4.8 1 3
4.6
0.5 2
4.4
Aα
AT
4.2 0 1
A
4
−0.5 0
3.8
3.6 −1 −1
0 10 20 30 0 10 20 30 0 10 20 30
T (s) T (s) T (s)
−0.21 −0.6 −0.3
−0.215 −0.61 −0.4

Aα̇
Aq
−0.22 −0.62 −0.5

b
−0.225 −0.63 −0.6
−0.23 −0.64 −0.7

0 10 20 30 0 10 20 30 0 10 20 30
T (s) T (s) T (s)
Figure 7: Load extraction simulation with fixed commands. Variation of the parameters in parameter vector A1
Also notice that only the fast dynamics parameters (α, θ, and q) change noticeably due to CG movement. The extrac-
tion time is was T E = 12.04s meaning it took approximately 7s for the load to travel from the CG to the rear of the
aircraft.
Adaptive controller - nominal case Figures 8 to 10 show the simulation results for the adaptive controller. The con-
troller parameters for this cenario are listed in table 3.
K 2
c1 1
P diag([1 1 1 1 1])
γ 1
Table 3: Controller parameters. Case 1.
It is seen the controller is effective in mantaining the pitch of the aircraft during movement of the load and after that
to reestablish the pitch after the load drop.
In figure 9 it is noted that the control effort does not impose any saturation problems, although it is certain that a more
realistic simulation would impose saturations in the elevator rate.
4522
ISSN 2176-5480
80 5
500
4
400 60
3
V (m/s)
H (m)
α (°)
300
40 2
200 1
20
100 0
0 0 −1
0 5 10 15 20 25 30 0 5 10 15 20 25 30 0 5 10 15 20 25 30
T (s) T (s) T (s)
2 4 5
3
1.5
2
∆Xcg (m)
q (°/s)
θ (°)
1 0
0.5
0
0
−1
−0.5 −2 −5
0 5 10 15 20 25 30 0 5 10 15 20 25 30 0 5 10 15 20 25 30
T (s) T (s) T (s)
Figure 8: Load extraction simulation with the adaptive controller (case 1). Evolution of the states.
5
3
δE (°)
−1
0 5 10 15 20 25 30
T (s)
Figure 9: Load extraction simulation with the adaptive controller (case 1). Elevator Command.
−3
x 10
15 0.015 3
10 2
0.01
Aα
AT
A0
5 1
0.005
0 0
−5 0 −1
0 5 10 15 20 25 30 0 5 10 15 20 25 30 0 5 10 15 20 25 30
T (s) T (s) T (s)
−0.05 −0.3 −0.3
−0.1 −0.4 −0.4

Aα̇
−0.15 −0.5 −0.5

A
−0.2 −0.6 −0.6
−0.25 −0.7 −0.7

0 5 10 15 20 25 30 0 5 10 15 20 25 30 0 5 10 15 20 25 30
T (s) T (s) T (s)
Figure 10: Load extraction simulation with the adaptive controller (case 1). Parameter variation and adaptation.
Yet perhaps the most surprising fact from the adaptation simulation is seen in figure 10. The results show a much more
effective adaptation of the paramters AT and A0 when compared to the adaptations of the other parameters. This suggests
that the gains in matrix P should be calibrated if one wants to favour the adaptation of all parameters.
Nevertheless, it can be considered that all the unmodelled dynamics were “absorbed” by the variation of those terms.
It is worth studying what would happen if those terms adaptations were shut down and the adaptations of the other terms
4523
ISSN 2176-5480

amplified. This is done in case 2.
6. CONCLUSIONS
This work has shown the feasibility of applying an adaptive controller to the problem of pitch regulation during a
payload extraction manouver. The controller design is nonlinear and based on the feedback linearization theory. The
adaptive architecture is a Model-Reference Adaptive System (MRAS or MRAC) and the resulting nonlinear-adaptive
controller was developed by direct application of Lyapunov’s second method.
Such a control strategy offer some advantages as the operation in more extreme deviations from the nominal flight
condition - thanks to the underlying nonlinear controller - and savings on the modelling effort, since a simpler model can
be used for the adpative case, as long as it takes into account all the main dynamic of the system.
There are also drawbacks to be pointed. The complexity of the controller is higher, as given by the augmented state
which encompasses the parameter’s estimates and the controller parameters such as the model reference and the controller
gains. Proofs of stability for the stochastic (real) case are still somewhat limited. Finally, as discussed in Astrom and
Wittenmark (2008), a robust linear controller may give better responses when fast parameter variation occurs.
Some sugestions for future work on the problem presented here are:
• The development of a least-square based, self-tuning controller for payload extraction , both in it’s direct and inderct
form, and a comparison of their performances with the MRAC studied in this work.
• Extending the application to the stochastic case, including sensor noise and other random disturbances on the system
and using the maximum likelyhood adaptive self-tuning controller for a LAPES (Low-Altitude Payload Extraction
System).
• Using the adaptive controller architectures to build a “auto-tunner” for simple UAV applications: Suppose an UAV
that works on a wide enough flight envelope to request different controller gains to keep the performance paremeters
under the various flight conditions. An adaptive controller could be used during the development phase in flight
tests to find the appropriate system gains, reducing the number of test flights, the control team workload and the
overall cost of the flight test campaign.
7. REFERENCES
Astrom, K. and Wittenmark, B., 2008. Adaptive Control. Dover, New York.
Duke, E., Antoniewicz, R. and Krambeer, K., 1988. “Derivation and definition of a linear aircraft model”. NASA Reference
Publication.
Etkin, B., 1972. Dynamics of Atmospheric Flight. John Wiley & Sons, New York.
Heffley, R. and Wayne, F., 1972. “Aircraft handling qualities data”. NASA Contractor Report NASA CR-2144, NASA.
Inc., T.M., 2000-2012. “Matlab”. http://www.mathworks.com/help/matlab/ref/odeset.html.
Roskam, J., 2001. Airplane Flight Dynamics and Automatic Flight Controls - Part I. Design, Analysis and Research
Corporation (DARcorporation), Lawrence, KS.
Rutan, E.L. and Stroup, F.B., 1967. “Evaluation of the c-130e sability and control characteristics during tandem, se-
quential, and single-paltform lapes delivery and airdrop deliveries”. Technical report, Air Force Test Center Edwards
AFB.
Singh, S. and Steinberg, M., 1996. “Adaptive control of feedback linearizable nonlinear systems with application to flight
control”. Journal of Guidance, Control, and Dynamics, Vol. 19, No. 4, pp. 871–877.
Steinberg, M.L. and Page, A., 1998. “Nonlinear flight control with a backstepping design aproach”. Professional paper,
Naval Warfare Center Aircraft Division.
Violato, G.O., 2012. Nonlinear Adaptive Control System For Payload Extraction Operations. Master’s thesis, Instituto
Tecnológico de Aeronáutica - ITA.
8. RESPONSIBILITY NOTICE
The author are the only responsible for the printed material included in this paper.
4524

Nonlinear, Adaptive Control System For Payload Extraction Operations

Uploaded by

Copyright:

Available Formats

Nonlinear, Adaptive Control System For Payload Extraction Operations

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Nonlinear, Adaptive Control System For Payload Extraction Operations

Uploaded by

Copyright:

Available Formats

ISSN 2176-5480

22nd International Congress of Mechanical Engineering (COBEM 2013)