You are on page 1of 7

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/290192997

Modeling and control strategy for the transition of a convertible tail-sitter UAV

Conference Paper · July 2007


DOI: 10.13140/RG.2.1.4925.9602

CITATIONS READS
15 236

4 authors, including:

Juan Escareno Anand Sanchez-Orta


XLIM Research Institute Center for Research and Advanced Studies of the National Polytechnic Institute
82 PUBLICATIONS 1,070 CITATIONS 119 PUBLICATIONS 1,747 CITATIONS

SEE PROFILE SEE PROFILE

R. Lozano
French National Centre for Scientific Research
682 PUBLICATIONS 19,427 CITATIONS

SEE PROFILE

All content following this page was uploaded by Anand Sanchez-Orta on 12 January 2016.

The user has requested enhancement of the downloaded file.


Proceedings of the European Control Conference 2007 WeC08.1
Kos, Greece, July 2-5, 2007

Modeling and Control Strategy for the Transition of a Convertible


Tail-sitter UAV
J. Escareño, R.H. Stone, A. Sanchez and R. Lozano

Abstract— This paper addresses the problem of the transi- two different flight regimes. In vertical flight thrust is the
tion between rotary-wing and fixed-wing flight of a tail-sitter dominant force and horizontal control is achieved via tilting
unmanned aerial vehicle (UAV). A nonlinear control design is the thrust vector, while maintaining attitude control using
presented to regulate the vertical-flight dynamics of the vehicle.
We present the dynamic and aerodynamic equations that model the propeller wash over the vehicle surfaces. (It should also
the behavior of the vehicle before (vertical flight), during and be noted that as an extra complicating factor, the amount
after (forward flight) the transition. A low-cost embedded of wash and hence the control effectiveness of the control
system, including an homemade inertial measurement unit surfaces depends on the thrust generated). In horizontal flight
(IMU), is used to perform autonomous attitude-stabilized flight the dominant force is the lift-force provided by the wings and
in vertical mode.
Index Terms— Tail-sitter, Backstepping, Embbeded architec- horizontal control is achieved via tilting this lift vector (in
ture, Inertial Measurement Unit. banked turns), while again maintaining attitude control with
the control surfaces in the free-stream flow. The design of a
I. INTRODUCTION robust control system that can cover these two flight regimes
and handle th non-linearities present in both is non-trivial
Tail-sitter UAVs have a number of advantages compared
task.
to other configurations. In comparison to conventional de-
In [1] the author utilize a LQR algorithm to control,
signs they poses much greater operational flexibility because
in hover mode, the longitudinal-flight dynamics (attitude
they don’t require a runway for launch and recovery but
and position). In [2] the authors apply a PD control to
instead can operate from any small clear space. While other
regulate the pitch and yaw attitude, with a classical fixed-
conventional designs partially overcome this limitation via
wing configuration, during vertical flight. In [3] we employ a
the use of takeoff and landing aids such as catapults and
saturation-based controller to stabilize the aircraft’s position
parachutes, these all entail extra system complexity and
and attitude at hover flight, assuming that it is close enough
logistic support. Although helicopter UAVs share the same
to the origin so that we could ignore the nonlinearity present
operational flexibility as the tail-sitter, they suffer from well
in the underactuated dynamics and also ignore the aerody-
known deficiencies in terms of range, endurance and forward
namics effects. This paper, however, considers the transition
speed limitations due to the lower efficiency of rotor-born,
between rotary-wing and fixed-wing flight, and thus the
rather than wing-born flight. Lastly, the other configurations
algorithm (backstepping-based controller) is required to be
that have been developed to archive the same goals as
robust enough to handle larger pitch angles away from the
the tail-sitter, such as the tilt-wing, tilt-rotor and tilt-body,
vertical. To do this the nonlinearity is explicitly accounted
do so at the expense of significantly increased mechanical
for, so that the UAV can achieve a vertical attitude from a
complexity compared to a tail-sitter that uses propeller wash
considerable pitch angle, (for instance during the transition
over normal aircraft control surfaces to effect vertical flight
from horizontal back to vertical flight).
control. By marrying the takeoff and landing capabilities of
The paper is organized as follows: in Section 2, the
the helicopter with the forward flight efficiencies of fixed-
mathematical model of the tail-sitter aircraft is presented. In
wing aircraft in such a simple way, the tail-sitter promises
Section 3, we develop a stabilizing control law for the vehicle
a unique blend of capabilities at lower cost than other UAV
in hover and forward flight mode. Simulations results are
configurations.
presented in Section 4. The experimental results are provided
While the tail-sitter concept has great promise, it also
in Section 5. Conclusions and perspective are finally given
comes with significant challenges. Foremost amongst these
en Section 6.
is designing controllers that will work over the complete
flight envelope of the vehicle: from low-speed vertical flight II. DYNAMICAL M ODEL
through to high-speed forward flight. The main change in this
The longitudinal model of vehicle that will be used in
respect (besides understanding the detailed aerodynamics) is
this paper is more complete than that used en previous
the large variation in the vehicle dynamics between these
papers about the tail-sitter [3]. This time simplified (though
J. Escareño and R. Lozano are with the HEUDI- reasonable) treatment of the aerodynamic lift, drag and
ASYC laboratory, Technology University of Compiegne, pitching moments will be given. It is important to consider
Compiegne, France. juan.escareno@hds.utc.fr, these forces properly because they are fundamentally affected
rogelio.lozano@hds.utc.fr
R.H Stone is with the School of Aeromech Engineering, University of by the vehicle’ s motion and thus alter the basic dynamics
Sydney, NSW, Australia. hstone@aeromech.usyd.au involved. The analysis used will be based on combination

ISBN: 978-960-89028-5-5 3385


WeC08.1

1 2
of a low-order panel method aerodynamic model coupled Cm = M/( ρVslip Sc ) (5)
with a simple actuator disc model of the flow induced by 2
the propellers. In considering the aerodynamic forces on the
vehicle it is essential to note that these are all associated
with the flow induced by the propellers over the vehicle’s In these equations S and c are a reference area and length
aerodynamic surfaces as well as by the perturbation of respectively. Thus to obtain the lift and moment forces on
this flow due to the motion of vehicle. Thus the starting the aircraft it is only necessary to obtain the slipstream axial
point for any analysis is determination of the propeller velocity, which has already been done, the angle of attack and
slipstream velocities. In the following analysis the following the aerodynamic parameters Clα , Clδ , Cmδ , Cmα which are
assumptions will be made: a function of the geometry of the vehicle. To determine the
A1. Propeller Normal Forces are negligible; angle of attack, it is necessary to obtain the W -component
A2. The slipstream velocity Vslip >> W , the body normal of relative velocity (the velocity component normal to the
velocity. This implies that the slipstream angle of attack is vehicle). This is given by:
small;
A3. The vehicle aerodynamic surface are fully submerged in
the propeller slipstream; W = Ẋ cos(θ) − Ż sin(θ) − Qlw (6)
A4. Body aerodynamic forces are negligible;
A5. Drag forces on the vehicle are small compared to thrust
and lift forces in hover flight, and small compared to lift
during horizontal flight. In this equation the first two terms come from the vehicle
linear motion, while the third is due to rotary motion about
Assuming purely axial flow into the propellers, (a close
the vehicle CG. This rotation at angular speed Q causes
enough assumption for the present case) simple actuator
an affective linear motion at the lifting surface, which is
disc theory [see [1]] gives the induced far-slipstream axial
displaced a distance lw from the CG. This treatment is not
velocity as:
s entirely accurate in the case where the aerodynamic surface
2T is large in comparison to lw however, this small error will
Vslip = U 2 + (1)
Aρ be ignored for simplicity. If the value of W is small in
comparison to the slipstream velocity, (true under the small
where U is the axial velocity of the vehicle, T , the thrust, a-assumption), then Vtot ≈ Vslip and the angle of attack can
A the total disc-area of the propellers and ρ the air-density. be written as:
This velocity must now be combined with the relative normal
velocity of the vehicle with respect ti the air to obtain the
aerodynamic forces. The relative normal velocity may be due à !
to vehicle motion (with respect to still air) or due to wind Ẋ cos(θ) − Ż sin(θ) − Qlw
αslip = arctan (7)
impinging on a stationary vehicle or a combination of these. Vslip
In the following derivation only the vehicle motion will be
considered with the addition of wind an obvious extension.
Fig. 1 shows the aerodynamic forces on a small UAV with its
lifting surfaces fully submerged in the propeller slipstream. T
The forces consist of a lift force, L, perpendicular to the
total flow vector, Vtot , a drag force parallel to Vtot , and a Propeller Disc
pitching moment, M , about the positive cartesian y axis, Z
.
a drag force, D, parallel. For small angles of attack, the
lw .
aerodynamic lift and moment forces will be proportional to Q X
the slipstream angle of attack, α, and to the deflection of the
control surface, δe . The drag force variation is more complex, L .
X Z
.
Vtot
however, it will be neglected for the rest of this treatment
Vtot
as it will usually be significantly less than the lift and thrust Vslip
terms. The above discussion can be summarized by:
M Relative Velocity
Component Build-up
Cl = Clα αslip + Clδ δ (2) D
at Aerodynamic
Qlw Reference Point
Z
Aerodynamic
Cm = Cmα αslip + Cmδ δ (3) Reference Point

where (2) and (3) are standard aerodynamic non-dimensional


X
lift and moment coefficients defined as:
1 2 Fig. 1. Aerodynamic forces: a UAV with lifting surfaces fully submerged
Cl = L/( ρVslip S) (4) in the propeller sliptream
2
3386
WeC08.1

A. Equations of Motion (Cartesian Form) It is clear that (11a) can be stabilized with an feedback-
The dynamical behavior of the tail-sitter vehicle [see figure linearizable input via the thrust T. An appropriate choice is
1] is described by the motion equations (8)1 rz + 1
T = (12)
 cos(θ)
 mẌ = T sin(θ) − L cos(θ + α) − D sin(θ + α)
mZ̈ = T cos(θ) + L sin(θ + α) − D cos(θ + α) − mg where, rz = −az1 Ż − az2 (Z − Z d ) with az1 , az2 > 0 and zd

Iyy θ̈ = −L cos(α)lw + M is the desired altitude. From (12) we notice that the controller
(8) scope is restricted to
In these equations the lift, drag and moment terms can be π π
− <θ<
represented as: 2 2
L = 21 ρS(Clα α + Clδe δe ) This interval is appropriate for vertical hover-mode, since
D = 12 ρS(Cd0 + KCL2 )VR2 the vehicle’s operating point usually is more or less the
M = 12 ρSc (CMδe δe )VR2 horizontal orientation. However, it excludes the possibility
for horizontal flight (θ = π2 ). This is entirely satisfactory as
where ρ is the density; S is the wing area submerged in the during horizontal flight the altitude dynamics are dominated
flow; and c is the wing-chord. by the wing lift force not the thrust force, which means that
B. Simplified model an alternative control strategy is required for that region of
the flight envelope.
For further control analysis let us consider a simplified Substituting (12) in (11) we obtain
model of (8). In this approximation we recall A1, A4 and A5, 
and we consider the normalized values of the mass, gravity,  Ẍ = (rz + 1) tan(θ)
the inertial mass moment (i.e. g = 1,m = 1, Iyy = 1). Z̈ = rz (13)
θ̈ = uθ (δe )

Therefore, (8) becomes
Note from (13b) that it exists a time large enough such that Z

 Ẍ = T sin(θ) − L cos(θ + α)

Z̈ = T cos(θ) + L sin(θ + α) − 1 (9) and Ż are arbitrarily small, hence x − θ dynamics becomes:
 θ̈ = 1 (−Le cos(α)lw )
½
Ẍ = tan(θ)

Iyy (14)
θ̈ = uθ (δe )
Since the lift provided by elevator deflection δe mainly acts
on the aircraft through the pitching moment, we may rewrite The state space representation of the previous equation is
θ̈ as: written 
ẋ1 = x2
θ̈ = uθ (δe ) (10)


ẋ2 = tan(x3 )

(15)
III. C ONTROL S TRATEGY 
 ẋ3 = x4
ẋ4 = uθ

In order to observe the aerodynamical behavior during
the vehicle’s transition we have defined the following task: is easy to observe that (15) has a pure-feedback system.
first, the UAV takes off vertically to a desired altitude z, Therefore, we will employ the Backstepping technique to
meanwhile it regulates the attitude θ to reach a desired design a control law that stabilize the underactuated subsys-
position x, afterwards, it switches to forward-flight modality tem.
controlling only the aircraft’s attitude. Firs step: Let us introduce a virtual state given by

A. Vertical Flight Control z2 = x2 − xv2 (16)


We consider the vertical flight as a critical stage for this at this step, we consider z2 = 0 to isolate (15a) and it also
vehicle, since our prototype’s structure is oriented more implies that x2 = xv2 . This shows that the first order system is
towards a classical fixed-wing aircraft. In this flight regime stabilisable through x2 . To do so, we propose the Lyapunov
exists two dynamic subsystems: the altitude dynamics, which function V1 = 12 x21 . Thus, to render V˙1 negative definite we
is fully-actuated by the thrust T , whereas the x−θ dynamics choose the following input
is an underactuated subsystem. Recalling A2, we neglect the
x2 = xv2 = −x1 (17)
aerodynamical terms in the control design. However, we will
show that our control law performs satisfactorily even if we which lead us to
include the aerodynamic term (perturbation) in the simulation V˙1 = −x21 (18)
study. The latter, allow us to rewrite (9) as:
Second step: Now, considering the case when z2 6= 0, the
second order system (15a-15b) becomes2

 Ẍ = T sin(θ)
Z̈ = T cos(θ) − 1 (11)
½
ẋ1a = z2 − x1

θ̈ = uθ (δe ) (19)
ż2 = tan(x3 ) + x2
1 In these equations we have dropped the subscript ”slip” from the angle 2 We use the subindex a to denote the augmented state variable which
of attack (α) contains the virtual state

3387
WeC08.1

TABLE I
At this step is convenient to introduce a second virtual state
S IMULATION ’ S PARAMETERS
given by
z3 = tan(x3 ) − xv3 (20)
PARAMETERS VALUE
considering z3 = 0 implies that tan(x3 ) = xv3 . Then, we m 1kg
use tan(x3 ) to stabilize the augmented second order system g 1m/s2
Iyy 0.01889kg · m2
(19). Note that, due to the tan(·) nature (several equilibrium
lw 0.3m
points), the controller acts within − π2 < x3 < π2 , which ρ 1.225kg/m3
is appropriate to perform vertical flight. We propose the Clα 1rad−1
Lyapunov function V2 = V1 + 21 z22 . Whose derivative V˙2 S 0.1653m2
is rendered negative definite through the following control Se 0.01m2
a θ1 10
tan(x3 ) = xv3 = −z2 − (x1 + x2 ) (21) a θ2 10
az1 1.5
leading V˙2 to az2 1
V˙2 = −x21 − z22 (22)
Third step: We consider the case when z3 6= 0, turning out
the following third order system B. Forward flight control

 ẋ1a = z2 − x1 The remaining stage in the assignment previously outlined,
ż2a = z3 − z2 − x1 (23) is the control of the pitch attitude during forward flight. Then,
ż3 = x4 tan2 (x3 ) + 2 tan(x3 ) + 2x2 to achive this goal we apply the following control input:

At this step is useful to introduce the third virtual state uθ = −aθ1 θ̇ − aθ2 (θ − θd ) (28)
z4 = x4 − xv4 (24)
with aθ1 , aθ2 > 0. At this stage, the convertible UAV is
with z4 = 0, we obtain x4 = xv4 . To derive the controller considered as a classical fixed-wing vehicle. The transla-
for (23) we propose the Lyapunov function V3 = V2 + 12 z32 tional behavior of the vehicle includes the altitude and the
whose derivative is rendered negative definite by horizontal motion, whose control inputs are, respectively, the
µ
z3 + z2 + 2 tan(x3 ) + 2x2
¶ elevator deflection and the thrust.
v
x4 = x4 = − (25)
tan2 (x3 ) + 1 IV. S IMULATION STUDY
Fourth step: Finally, we consider z4 6= 0, turning out the In order to validate the control strategy described in
following system Section 3, we have run simulations (see table I) to observe


 ẋ1a = z2 − x1 the performance of the aircraft in both flight regimes. The
 ż = z − z − x
2a 3 2 1 UAV starts performing hover flight from the initial position
2 (26)
 ż3a = z4 (tan (x 3 ) + 1) − z3 − z2 (x, z, θ) = (0, 0, π8 ), then the vehicle reaches a desired
ǫ1

 ż = u +
4 θ tan2 x3 +1 altitude (z = 2) with desired position (x = 0.6) and pitch
with attitude (θd = 0), then, at the time t = 25 a perturbation
(θp = π8 )) in the attitude is applied. At the time t = 50
ǫ1 = 3x2 + 5 tan x3 + 3x4 (1 − tan2 x3 )
the aircraft switches to forward flight, considering θd = π4 .
−2x4 tan x3 (3x1 + 5x2 )
Finally, at the time t = 60 a perturbation π8 is applied to the
In order to deduce the controller for (26) we propose the pitch attitude.
Lyapunov function V4 = V3 + 12 z42 , whose derivative is The simulation corresponds to the system described by
definite negative as long as we apply the following the equations (9). Therefore, we notice in the figure (2) that
ǫ2 the hover-mode controllers (12) and (27) are robust enough
uθ = −z4 − (27)
tan2 x3 + 1 to cope with the aerodynamical perturbation (adverse lift)
with created by a considerable vertical position deviation.
ǫ2 = 2x1 + 5x2 + 6 tan x3 + 3x4 (1 − tan2 x3 ) In figure (3) we observe that the elevator deflection
−2x4 tan x3 (3x1 − 5x2 ) presents an overdamped behavior, mainly due to the ele-
vator’s operation interval and the unitary gains used in the
The virtual states obtained along the controller design are
backstepping algorithm. Moreover, we can observe the thrust
given by:
z2 = x2 + x1 performance in both modalities. In figure (4) we observe the
z3 = tan(x3 ) + 2x1 + 2x2 variation of the lift during the presence of a perturbation in
the pitch attitude. After the transition we observe the relation
z4 = x4 + 3x1 +5x2 +3 tan(x3 )
tan2 (x3 )+1 between the lift-force the attack angle.
The final control law (27) stabilizes the underactuated sub-
system (15), at hover mode, in attitude and position. It is V. E XPERIMENTAL RESULTS
worth to remark that we have used unitary gains during the The configuration of the tail-sitter UAV is a compromise
backstepping design to avoid the abuse in the notation. between rotary and fixed wing aircraft. Hence, the flight

3388
WeC08.1

1.5 2

θ velocity [rad/s]
1 1 1 1.5

Lift coefficient
θ [rad]

α [rad]
0.5 1
0 0
0 0.5
−1 −1
−0.5 0
0 20 40 60 80 0 20 40 60 80 −1 −0.5
time[s] time[s] 0 20 40 60 80 0 20 40 60 80
α [rad] time[s]
3 1.5

x velocity [m/s]
0.4 10
x position [m]

Slipstream velocity [m/s]


2
0.5 8
1 0.2
0

Lift [N]
6
0 −0.5 0
4
−1 −1
0 20 40 60 80 0 20 40 60 80 −0.2
2
time[s] time[s]
8 2 −0.4 0
0 20 40 60 80 0 20 40 60 80
z velocity [m/s]

time[s] time[s]
z altitude [m]

6
1
4
0 Fig. 4. UAV’s aerodynamics
2

0 −1
0 20 40 60 80 0 20 40 60 80
time[s] time[s]
to the Rabbit microcontroller throught the serial port.
The inertial information is filtered to get rid of the
Fig. 2. UAV’s states electrical and mechanical noise (mainly due to rotor’s
gearbox). Finally, the control signal is sent to the
2 1.5 motors (propellers) and servomotors (ailerons-elevator)
1
via the PWM ports (see fig. 7).
Thrust [N]

1
δ [rad]

0 B. Attitude performance
e

0.5
−1 So far, we have performed experimental test, on the
−2 0 prototype previously described, to stabilize the aircraft’s
0 20 40 60 80 0 20 40 60 80
time[s] time[s] attitude and in vertical flight mode. The performance of this
experiment is shown in figures 8, 9 and 10. y
Fig. 3. UAV’s control inputs
VI. C ONCLUDING REMARKS
In this paper we have presented a control strategy to
control of the UAV, in both operations modes, depends on perform the transition of a tail-sitter UAV from vertical to
dynamic and aerodynamic terms. The altitude of vehicle is forward flight mode. We have presented in detail the longitu-
regulated by increasing or decreasing the propeller thrust. dinal model of the aircraft as well its simulation in order to
The roll torque is obtained from the difference of the rotors observe the dynamic and aerodynamic performance during
angular velocities. Since the control surfaces are submerged the transition. For vertical flight mode, we have derived
in the propeller slipstream (prop-wash), the aerodynamic control algorithm, based on backstepping, robust enough
forces are generated by the elevator and ailerons deflections to cope with considerable aerodynamical perturbations. An
to provide the pitch and yaw motion, respectively. embedded autopilot was successfully tested for the attitude
stabilization in vertical flight.
A. Embedded system
In this section we described the embedded system to
perform an autonomous attitude-stabilized flight. The system
is composed by two main modules: the inertial measurement
unit (IMU), the embedded control module.
1) Intertial Measurement Unit: We have built an inertial
measurement unit (IMU) that includes a dual-axis
accelerometer, which senses the angular position of
the vehicle (φ, θ), and three gyros, which sense the
angular rate of the vehicle (φ̇, θ̇, ψ̇). The yaw 3 angle
is obtained by the integration of yaw angular rate.
2) Embedded Control: The IMU (analog signals) feeds
the PIC microcontroller which sends this information
3 The associated drift is tackled by resetting the yaw angle at a certain
drift value Fig. 5. VTOL prototype

3389
WeC08.1
5

imu.jpg
4

Roll angle [°]


0

−1

Time [s]
−2

−3

−4

−5
0 20 40 60 80 100 120 140

Fig. 8. Roll Angle performance


5
Fig. 6. Homemade IMU
4

3
IMU
GYRO Z GYRO X GYRO Y 2

Pitch angle [°]


1

ACC X ACC Y 0

−1

Time [s]
−2
ADC PIC
RC Receiver −3
SERIAL

−4
PWM
PING

−5
0 20 40 60 80 100 120 140

Fig. 9. Pitch Angle performance


D-Fusion PWM 50
Capture 40

30

20

Σ
Yaw angle [°]

10

Control 0
PWM

−10
Rabbit
−20

−30

−40

−50
0 20 40 60 80 100 120 140
Time [s]
C-surfaces Propeller
Actuators
Fig. 10. Yaw Angle performance
Fig. 7. Homemade IMU

[5] H. Goldstein, C.P. Poole and J.L Safko, Classical Mechanics, Addison-
The challenge is now to expand the control strategy Wesley Publishing Company, Inc., Massachusetts, 1983.
[6] A. Bedford, and W. Fowler, Dynamics, Addison-Wesley Publishing
for the 6-DOF nonlinear model of the tailsitter UAV for Company, 1989.
both operation regimes. Moreover, the improvement of the [7] P. Castillo, R. Lozano A. Dzul, ”Modeling and control of mini flying
homemade IMU in such a way that it can be capable to machines, Springer-Verlag, July 2005.
[8] R. Lozano, et. al, Dissipative Systems Analysis and Control, Springer-
measure the inertial attitude in the whole vehicle’s operation Verlag, 2001
range, as well the incorporation of sensors such as GPS or [9] I. Fantoni and R. Lozano, Nonlinear Control for Underactuated
vision-based sensors. Mechanical Systems, Springer-Verlang, 2002.
[10] B. Etkin and L. Reid, Dynamics of Flight, J. Wiley & Sons, Inc., 1991.
[11] B. L. Stevens and F.L. Lewis, Aircraft Control and Simulation 2ed.,
R EFERENCES J. Wiley & Sons, Inc., 2003.
[1] R. Hugh Stone, ”Control Architecture for a Tail-sitter Unmanned Air
Vehicle”, 5th Asian Control Conference, Melbourne, Australia, July
23-25, 2004.
[2] William E. Green and Paul Y. Oh, ”Autonomous Hovering of a Fixed-
Wing Micro Air Vehicle”, International Conference on Robotics and
Automation, Orlando, Florida, USA, May, 2006.
[3] J. Escareno, S. Salazar and R. Lozano, ”Modeling and Control of a
Convertible VTOL Aircraft”, 45th IEEE Conference on Decision and
Control, San Diego, California, December 13-15, 2006.
[4] H. Stone, ”Aerodynamic Modeling of a Wing-in-Slipstream Tail-Sitter
UAV”, Biennial International Powered Lift Conference and Exhibit,
Williamsburg, Virginia, Nov. 5-7, 2002.

3390

View publication stats

You might also like