LONGITUDINAL CONTROL OF A FIXED WING UAV

David Villota Miranda, Montserrat Gil-Martínez, Javier Rico-Azagra

Control Engineering Group, Electrical Engineering Department, University of La Rioja
[email protected], {montse.gil, javier.rico}@unirioja.es

Summary the worst part, as the connection to the main body is

weak.
This article presents an approach to the mathematical
model of a fixed wing unmanned aerial vehicle Regarding control, it is challenging since, in contrast
prototype. The model is split in two different parts, to the conventional architecture, where there are
related to the longitudinal and lateral stability, multiple control surfaces, in a fixed wing aircraft we
respectively. For this, Newton-Euler formulation is can only control the flap deflection in a symmetric or
used as well as basic aerodynamic theory. asymmetric way, depending which stability
Aerodynamic coefficients, inertias and characteristic (longitudinal or lateral) is to be controlled.
points of the aircraft are obtained through simulations
with an open-source software called XFLR-5, and the For this work, an unconventional plane (fixed wing
physical parameters of the model match the architecture) has been built with all the required
prototype’s. Then, a longitudinal control strategy avionics and actuators (servo motors, Ardupilot,
describes the altitude control in a cascade ESC,…). Figure 1 illustrates its dimensional
architecture, whose inner loop conveniently characteristics. The final aim is the development of a
manoeuvres the pitch angle by acting on the navigation and control system not to depend on
symmetric flag deflection. Frequency domain commercial solutions. As a preliminary work, this
techniques are used to design PID controllers. paper describes the mathematical modelling of the
fixed wing prototype, and focuses on the linear model
Key words: Fixed-wing unmanned aerial vehicle for longitudinal stability. Newton-Euler formulation
(UAV), stability derivatives, cascade control, [3] is being used, but the mayor difficulty is the
proportional integral derivative control. identification of aerodynamically coefficients and
characteristics points that will be obtained through
simulations with an open-source software called
XFLR-5 [4]. Then, a control strategy is being
presented in order to firstly control the pitch angle,
which modifies the angle of attack that modifies the
1. Introduction magnitude of the lift force. This ultimately leads to
proper altitude control. The system to be controlled is
The unmanned aerial vehicles (UAV) are becoming multivariable, non-linear and highly coupled.
more and more popular as they can be used in a wide
range of fields. The technology involved is being
continuously developed and its price is constantly
decreasing [1]. Every time open-source projects are
being carried out by research groups and more and
more information can be found on the internet, given
by Radio Control fan communities.

In the UAV field, of the four different categories, it is

micro and mini that have experimented a greater
expansion. Even though multirotors are the most
popular, glider UAVs have more advantages as their
autonomy is greater. That is why they are used in field Figure 1: Aircraft prototype
recognition or to reach further places that multi-rotors
cannot [2]. 2. Fixed wing UAV modelling
The fixed wing architecture has been chosen as it is UAV movement is defined by 6 degrees of freedom
easier to maintain than a conventional plane, and more (DoF) in the earth inertial frame (E): 3 coordinates for
resistant in case of crushing. Normally the tail suffer spatial positioning [x,y,z]T and 3 angles for orientation
[ , , ] T. A fixed wing is an unconventional type of
aircraft with only two control surfaces (flaps) for 𝐹𝑥 = 𝑚(𝑢̇ + 𝑞𝑤 − 𝑟𝑣)
manoeuvring (Figure 2). The symmetric deflection ( {𝐹𝑦 = 𝑚(𝑣̇ + 𝑟𝑢 − 𝑤𝑞) ( 1)
 E ) of flaps will act in the control of the longitudinal 𝐹𝑧 = 𝑚(𝑤̇ + 𝑝𝑣 − 𝑢𝑞)
motion variables (  , z ), and the asymmetric
where m is the plane mass.
deflection (  A ) of flaps will act in the control of the
lateral motion variables (  , ). The controlled Euler’s second law on the angular momentum yields
attitude (, ), and heading () are further responsible the contribution of the three torque components in the
of spatial displacement. body frame:

𝑀𝑥 = 𝐼𝑥𝑥 𝑃̇ + (𝐼𝑥𝑥 − 𝐼𝑦𝑦 )𝑄𝑅 + 𝐼𝑥𝑧 𝑃̇ + 𝐼𝑥𝑧 𝑃𝑄

{ 𝑀𝑦 = 𝐼𝑦𝑦 𝑄̇ + (𝐼𝑥𝑥 − 𝐼𝑧𝑧 )𝑃𝑅 + (𝑅2 + 𝑃2 )𝐼𝑥𝑧 , (2)
𝑀𝑧 = 𝐼𝑧𝑧 𝑅̇ + (𝐼𝑦𝑦 − 𝐼𝑥𝑥 )𝑃𝑄 + 𝐼𝑥𝑧 𝑃̇ − 𝐼𝑥𝑧 𝑄𝑅

being the inertial tensor:

Figure 2: Symmetric  E and asymmetric  A
deflections  I xx 0 I xz 
I   0 I yy 0  (3)
Linear [u, v, w]T and angular [ p, q, r ]T velocities of  I zx 0 I zz 
the fixed wing are measured in the body frame (B).
Figure 3 depicts the three orthogonal axis of this Mass and inertial moments for the fixed wing
second reference frame, which is clamped to the mass prototype in this work are in Table 1, together with
centre of the vehicle. other relevant parameters. Moments of inertia have
been computed using the 3D simulation program
In all aerodynamic systems, special attention should XFLR-5 [4].
be payed to the “wind frame” (W), whose X-axis is
parallel to the air velocity vector Va . W reference Parameters Values Units
frame involves a rotation  (attack angle) with respect m 0.9 𝑘𝑔
to the body Y-axis and a rotation  (sweep angle) with A 0.27 𝑚2
respect to the body Z-axis, as Figure 3 illustrates. Va b 1 𝑚
magnitude depends on the relative vehicle’s forward c 0.27 m
airspeed. Va 31 𝑚/𝑠
 −0.5 º
 0 º
g −9.81 𝑚/𝑠 2
kd 8.5 ∗ 10−9 𝑁 ∙ 𝑚 ∙ 𝑠 2 /𝑟𝑎𝑑
kt 5.65 ∗ 10−7 𝑁 ∙ 𝑚 ∙ 𝑠/𝑟𝑎𝑑
Ixx 0.02381 𝑘𝑔 ∙ 𝑚2
Iyy 0.00841 𝑘𝑔 ∙ 𝑚2
𝑋𝐵 Izz 0.03222 𝑘𝑔 ∙ 𝑚2
𝑌𝑤 𝛼 Ixz=-Izx 0 𝑘𝑔 ∙ 𝑚2
𝛽 𝑰𝑿𝒋 2.44 ∗ 10−6 𝑘𝑔 ∙ 𝑚2
𝑉𝑎
𝑌𝐵 𝑍𝑤 Table 1. Parameters of fixed wing prototype and
𝑍𝐵 flight conditions
𝑋𝑤

Figure 3: Body clamped frame (colour) and wind 2.2 External strengths and torques
frame (grey)
The relative vehicle’s forward airspeed Va exerts an
2.1 Equations of motion in the body frame aerodynamic strength due to the variance of pressure

Newton’s second law on the linear momentum yields 𝑄 = 12𝜌𝑉𝑎2 𝐴 (4)

the contribution of the three force components in the
body frame:
between the upper and lower parts of the wing, whose propeller 5051, all powered with a 4S LiPo battery.
surface is A;  is the air density. Thus, the drift (D), Propellers coefficients k t and k d in Table 1 have been
sweep (S) and lift (L) components of the aerodynamic experimentally identified according the procedure in
strength in the wind frame are [5].
−𝐷 −𝐶𝐷
The propeller rotation axis changes its orientation as
𝐅𝑎W = [ 𝑆 ] = 12𝜌𝑉𝑎2 𝐴 [ 𝐶𝑆 ], (5) the craft rotates. This induces a gyroscopic torque
−𝐿 −𝐶𝐿
𝐼𝑗𝑋 𝜔̇ 𝑗
being 𝐶𝐷 , 𝐶𝑆 , and 𝐶𝐿 the aerodynamic coefficients in
each W axis. Then, the rotation matrix 𝐌𝒈 = [ 𝐼𝑗𝑋 𝜔𝑗 𝑟 ] (11)
−𝐼𝑗𝑋 𝜔𝑗 𝑞
cosα cosβ −cosα sinβ −sinβ
B←W
𝐑 = ( sin β cos β 0 ) ( 6) where 𝐼𝑗𝑋 is the moment of inertia of the rotor around
sin α cos β −sinα cos β cosα the X-body axe.

is applied to obtain those strengths in the body frame: Finally, the craft weight in the earth frame responds to
0
𝐹𝑎𝑋 𝐅𝒘𝐄 = [ 0 ] (12)
𝐅𝑎 = 𝐑 B←W
𝐅𝑎W = [𝐹𝑎𝑌 ] ( 7) −𝑚 𝑔
𝐹𝑎𝑍
where m is the mass of the aircraft and g is the gravity.
The application point of 𝐅𝑎 can slightly change Then, this force is conveniently rotated to the body
depending on the attack  and sweep  angles. In frame giving
order to simplify the problem, the application point is
−𝑚 𝑔 𝑠𝑖𝑛𝜃
considered fixed and roll (L), pitch (M) and yaw (N)
moments 𝐅𝒘 = [ 𝑠𝑖𝑛𝜙 𝑐𝑜𝑠𝜃 ]
𝑔 (13)
𝑔 𝑐𝑜𝑠𝜙 𝑐𝑜𝑠𝜃
𝐿 −𝑏 𝐶𝑙
𝐌𝒂 = [𝑀] = 12𝜌𝑉𝑎2 𝐴 [ 𝑐 𝐶𝑚 ] (8) 2.3 Non-linear model
𝑁 −𝑏 𝐶𝑛
Substituting external forces and moments (Section
are added to correct this assumption; b and c are the 2.2) in generic forces and moments
wing span and chord, respectively; Table 1 details Fx , Fy , Fz , M x , M y , M z in (1) and (2), and
their values for this work prototype. 𝐶𝑙 , 𝐶𝑚 , 𝐶𝑛 are rearranging, it yields the dynamic non-linear model of
aerodynamic coefficients in each axis. They depend motion in the body frame:
on the attack angle (), the flap deflection (  E ,  A )
2
𝑘𝑡 (𝑤𝑗 ) 𝐹𝑎𝑋
and the angular velocities [ p, q, r ]T . Translational (5) −𝑔 𝑠𝑖𝑛𝜃 + − + 𝑣𝑟 − 𝑞𝑤 = 𝑢̇
𝑚 𝑚
and rotational (8) aerodynamic coefficients have been 𝐹𝑎𝑌
calculated following the equations in [3]. 𝑔 𝑠𝑖𝑛𝜙 𝑐𝑜𝑠𝜃 + + 𝑝𝑤 − 𝑟𝑢 = 𝑣̇
𝑚
𝐹𝑎𝑍
𝑔𝑐𝑜𝑠𝜙 𝑐𝑜𝑠𝜃 − + 𝑢𝑞 − 𝑣𝑝 = 𝑤̇
A tail propeller rotates at j, which provides a thrust 2 1
𝑚
𝐼𝑗𝑋 𝜔̇𝑗 𝑘𝑑 (𝑤𝑗 ) 𝜌𝑉 2 𝑆𝑐𝑙 𝑏 (𝐼𝑥𝑥 −𝐼𝑧𝑧 )𝑞𝑟 (14)
force along the X-body axis + − 2
− = 𝑝̇
𝐼𝑥𝑥 𝐼𝑥𝑥 𝐼𝑥𝑥 𝐼𝑥𝑥
1
𝐼𝑗𝑋 𝜔𝑗 𝑟 𝜌𝑉 2 𝑆𝑐𝑚 𝑐
T j = k t (  j )2 (9) +2 −
(𝐼𝑧𝑧 −𝐼𝑥𝑥 )𝑝𝑟
= 𝑞̇
𝐼𝑦𝑦 𝐼𝑦𝑦 𝐼𝑦𝑦
1
−𝐼𝑗𝑋 𝜔𝑗 𝑞 𝜌𝑉 2 𝑆𝑐𝑛 𝑏 (𝐼𝑥𝑥 −𝐼𝑦𝑦 )𝑝𝑞
to get the plane sustentation force. However, the 2
{ − − = 𝑟̇
𝐼𝑧𝑧 𝐼𝑧𝑧 𝐼𝑧𝑧
friction between the propeller and the air also causes a
parasitical drag moment around the X-body axis
The linear velocities [u, v, w]T can be transferred to the
earth frame by multiplying them by matrix (see [5])
 j = k d (  j )2 , (10)
𝐑E←B = 𝐑E←B
𝑥 ∙ 𝐑E←B
𝑦 ∙ 𝐑E←B
𝑧 , (15)
which hampers the plane controllability. Thus, it is
worth investing time to find the best motor-propeller And after integration, it yields absolute position [x, y
combination. For this work prototype, we have opted z]T. Expression (15) uses Euler angles that can be
for a motor Racestar BR2205, 2300Kv, with a 3-blade calculated integrating
 𝛼(𝑠) −10.95 𝑠3 −2964 𝑠2 −569.6 𝑠−334.3
= 𝑠4 +30.37𝑠3 +3987𝑠2 +787.8 𝑠+658.7 (20)
𝜙̇ 1 𝑠𝑖𝑛 𝜙 𝑡𝑎𝑛𝜃 𝑐𝑜𝑠 𝜙 𝑡𝑎𝑛𝜃 𝑝̇ 𝛿𝐸 (𝑠)
[ 𝜃̇ ] = [0 𝑐𝑜𝑠 𝜙 − 𝑠𝑖𝑛 𝜙 ] ∙ [𝑞̇ ] (16)
are of interest in the longitudinal control strategies.
𝜓̇ 0 𝑠𝑖𝑛 𝜙 𝑠𝑒𝑐 𝜃 𝑐𝑜𝑠 𝜙 𝑠𝑒𝑐𝜃 𝑟

This also yields absolute orientation [ , , ] T.

3. Control strategy
In flight dynamics, it is all about controlling the
2.4 Linear model: Longitudinal stability
magnitude and orientation of the lift vector. Thus, we
will have to study those variables whose effect on the
In order to develop linear control laws, small signal
vector are significant. We find that the attack angle
linear models of (14) will be computed. The
controls the magnitude and the sweep angle controls
linearization process is about deriving the equations
the orientation of the lift vector. Consequently, any
regarding all variant parameters, evaluating them on a
manoeuvre of winning or losing height would start
nominal flight condition (Va=31 m/s, =-0.5º , =0º),
with a change in the angle of attack, in the same way
and multiplying them by the sensitivity. The result of
that a change in the sweep angle has an inherited
this process is commonly called stability derivatives in
change in the lateral position.
the aeronautic field. Longitudinal stability is used for
pitch and height control, and lateral stability for roll
The desired attack angle is obtained by controlling a
and yaw control [3] [6].
desired pitch angle, which finally will intervene in the
altitude control. Similarly, a desired sweep angle is
The longitudinal stability allows us to observe the
obtained by controlling a desired roll angle, which will
behaviour of the linear velocity in X-axis (𝑢), the intervene in the yaw control.
angle of attack (), the angular velocity in Y-axis (𝑞)
-all them computed in the body frame- , and the pitch In this work, two cascaded loops will allow controlling
angle (  ) in the earth frame, which is approximated first the pitch angle. Then, considering the attack
by the integration of the aforementioned angular angle, it will allow controlling the height inside
velocity q under the assumption of small roll angles. another outer loop.
The symmetric deflection of flaps  E is the actuation
variable. 3.1 Pitch control architecture

Only the strengths in X and Z axes, and the moments Figure 4 depicts the pitch control architecture. Block
in Y axis will be studied, since they are the only ones a(s) represents the actuator dynamic, which is here
deemed to intervene in longitudinal stability. With the discarded (a(s)=1) in comparison with the rigid solid
coefficients obtained following [7], it yields the dynamics (s)/E(s). The pure derivative in the inner
longitudinal linear model: loop is actually a mathematical resource, since q is the
measurable variable in practice. Thus, gain Kq is the
𝑋𝑢 + 𝑋𝑇𝑈 𝑋𝑎 0 −𝑔 𝑋𝛿𝐸 controller in the feedback path of the inner loop. The
𝑢̇ 𝑍𝑞 + 𝑈0
𝑢
𝑍𝑢 𝑍𝛼 𝑍 𝛿𝐸
𝛼̇ 0 𝛼 outer loop provides the feedback controller C(s) in
( )= 𝑈0 𝑈0 𝑈0 (𝑞 ) + 𝑈0 𝛿𝐸
𝑞̇ the direct path. The control design process is
𝑀𝑢 + 𝑀𝑇𝑈 𝑀𝛼 + 𝑀𝑇𝛼 𝑀𝑞 0 𝑀𝛿𝐸
𝜃̇ 𝜃
performed from the inner to the outer loop, as it is
( 0 0 1 0) ( 0 )
(17) following detailed.

In particular, the state equation

𝑢̇ −0.19 0 0 −9.17 u 0.51

𝛼̇ 0 −20.27 0.96 0 α −11.05
( )=( ) (q ) + ( ) 𝛿𝐸
𝑞̇ 4.04 −3861.27−9.90 0 −2985.33
𝜃̇ 0 0 1.00 0 θ 0
(18)

is obtained for the fixed wind prototype in this work. Figure 4: Pitch control architecture
Accordingly, the following input-output transfer
functions From a pure mathematical point of view the
diferenciator in the inner loop mitigates the under-
𝜃(𝑠) −2985 𝑠2 −18820 𝑠−3487
= (19) damping (0.244) of dominant poles in (s)/E(s) of
𝛿𝐸 (𝑠) 𝑠4 +30.37𝑠3 +3987𝑠2 +787.8 𝑠+658.7
(19). Figure 5 depicts this effect in the frequency
domain response of /E. Then, Kq is tuned to achieve
a suitable control bandwidth BW; acceptable values
are between 1 and 10 rad/s. Finally, a value of which achieves a PM of 90º at gc of 9.78 rad/s, as
Figure 7 depicts. Finally, the closed-loop frequency
𝐾𝑞 = −0.25 (21) response /c reaches -3dB above BW=5.5 rad/s.

achieves a BW=1.71 rad/s, as Figure 6 shows. Let us

remark that a negative control gain is necessary in the
inner loop since (s)/E has inverse gain.

Figure 7: Open-loop frequency response of /e

3.2 Altitude control architecture

Figure 5: Open-loop frequency response /E The altitude control consists of another feedback
control loop above the pitch control structure /c of
Figure 4, as Figure 8 shows. C h (s) is the feedback
controller to be designed. The path angle

 (t )   (t )   (t ) (23)

is related to the altitude such that


h  U 0 sin   U 0 , (24)

where U0 is the craft velocity that is equal to Va=31

m/s when ==0º.

Figure 6: Closed loop frequency response /c

Regarding the outer loop design, a proportional- Figure 8: Altitude control architecture
integral (PI) controller is attempted: first, an integrator Considering (23) (19) and (20), it is obtained
to remove the position error and later on a zero to
mitigate the integrator effect over medium frequencies  (s) -0.003667( s + 39.77)(s - 41.91)(s + 0.1727)
guaranteeing enough phase margin (PM) -higher than = ,(25)
 (s) (s + 6.115)(s + 0.191)
40º-. The PI controller gain modulates the gain cross
over frequency gc (values between 1 and 10 rad/s are
acceptable). Negative control gain is necessary since which can be approximated by
/c has inverse gain as phase plot reveals in Figure 6.
(s) 5.4189
The final controller at the outer loop is  (26)
(s) s  6.115
(1 + 5s)
C (s) = -0.632
s (22)
in order to simplify the design process. Finally, h/c with the symmetric and asymmetric flap deflection as
presents the frequency response in Figure 9. control inputs, and the three Euler Angles and altitude
as controlled outputs. Using this block, the
A PI controller cannot achieve acceptable PM (above aforementioned control loops (Sections 3.1 and 3.2)
45º) for good stability and high enough cross over- have been also implemented in the script. Besides,
frequencies (1-10 rad/s) for a good performance. yaw must be controlled to zero, using a similar control
However, the final controller structure (Figure 11) as in the height control, let us
note as it includes inner roll control loops, similar to
0.05 (1  s / 0.9)( s  1) the pitch control architecture.
Ch ( s)  (27)
s (1  s / 2.4)

achieves a PM of 51.45º and cg=2.54 rad/s as Figure

10 shows. The closed-loop bandwidth of h/hc is
BW=4.25 rad/s.
Figure 11: Yaw control architecture

Figure 12 depicts several time responses related to

height reference changes of step and ramp type. Plot
(a) depicts the height tracking response (black) to
reference signals (grey) of different nature; plot (b)
shows the pitch that is demanded (grey) and how it is
attained (black) by the inner loop; and plot (c) shows
the deflection angle variation.

Figure 9: Frequency response of h/c

Figure 10: Open-loop frequency response of h/eh

Figure 12: Altitude control performance

3.3. Validation in the non-linear model

4. Conclusions
Longitudinal control is being tested in the fully
coupled system with all the non-linear behaviours. In this article, we have presented the mathematical
The non-linear model in Section 2.3 has been model of a fixed wing aircraft. For a hand-made
implemented in a “User-Defined Block” in Simulink protype, we have identified aerodynamic and physical
parameters such as aerodynamic coefficients, inertias References
or weights, among others, mainly using the open-
source software XFLR-5. Furthermore, we have [1] Mark Edward Peterson. (2006) “The UAV and
isolated all the moments and strengths in the system: the current and future regulatory construct for
weight, aerodynamic forces and moments, thrust, drag integration into the national airspace system”.
sweep. Journal of Air Law and Commerce, vol. 71(5)
pp. 521-612.
Following Newton-Euler formulation, we have come
up with a non-linear model, which has been linearized [2] Alexander V. Koldaev. (2007) “Non-military
in order to apply linear control theory. UAV applications”.

A longitudinal stability model has been used to design [3] David K. Schmidt. (2012) “Modern Flight
feedback control loops of a cascade structure. Dynamics”, McGraw-Hill International Edition.
Frequency domain techniques were used to design
PID type controllers. An inner feedback loop [4] Guidelines for XFLR5 V0.03. (2009) “XFLR5
controlled the pitch angle by conveniently acting on Analysis of foils and wings operating at low
the flap deflection. Then, an outer loop allowed Reynolds numbers”.
tracking the desired altitude.
[5] R.Rico, P. Maisterra, M. Gil-Martínez, J. Rico-
Achieving this controlled model is the start of a way Azagra, S. Nájera (2015). “Identificación
for improvement and allows us to contribute to the experimental de los parámetros de un
creation of navigation systems, laying the foundations cuatrirrotor”. In XXXVI Jornadas de
of new work lines. Automática.

The development of the model and its control is the [6] Smetana,Frederick O. Delbert C. Summery and
first step to design optimized control strategies and to W. Donlad Johnson (1972), “Riding and
explore new possibilities in the field. Handling Qualities of light Aircraft-A Review
and Analysis”. National aeronautics and space
administration. Washington D.C

Acknowledgements [7] Esteban. S., (2001) “Static and dynamic analysis

of an unconventional plane: Flying wing”. In
The authors gratefully appreciate the support given by AIAA Atmospheric Flight Mechanics
La Rioja Government under grant ADER 2017-I-IDD- Conference and Exhibit.
00035 and the support given by the University of La
Rioja under grant REGI 2018/42.