Modelling the dynamics of an unmanned aircraft using a graphical

simulation tool
Sara Jansson Ahmed

Department of Aeronautical and Vehicle Engineering

KTH - Royal Institute of Technology

SE 100 44 Stockholm, Sweden

2012-03-12

Complex systems can be modelled using graphical tools such as Simulink. Here, the equa-
tions of motion in combination with the necessary data were modelled using Simulink to de-
scribe and analyze the unmanned Predator C Avenger. The aerodynamic data was obtained
from the potential ow solver dwfs (dry water ow solver). The Simulink model is initially in
steady state, disturbances are then added to investigate the ying qualities of the aircraft both
in the longitudinal and lateral direction. The modes calculated from the model are compared
to the modes produced from the linearized matrices containing the aerodynamic derivatives.
The other task for the Simulink model is for some of its subsystems to be integrated with the
simulation program ARES, which is used at Saab. The Simulink subsystems are converted
into C code and are built into ARES. The rst expected result is that the modes from the
linearized model should indicate the values of the modes from the Simulink model. The largest
dierence - which regards the spiral mode - was about 24% and assumed acceptable as the
approximative solution for this mode is considered poor [4]. The other dierences were at
most about 15%. The second expected result is that the outputs from simulating the complete
model using Simulink should match the outputs obtained from simulating through ARES con-
nected to the Simulink subsystems. It was the discovered that the obtained results - which are
the values of the longitudinal trim parameters and the modes - from Simulink match almost
exactly the results from ARES with some few deviations. The trim results had a maximum
dierence of about 6% and the values of the modes had a maximum dierence of about 10%
and also assumed acceptable.

1 Nomenclature
2
α = angle of attack [rad] Ixx = mass moment of inertia around the x-axis [kgm ]
2
β = sideslip angle [rad] Iyy = mass moment of inertia around the y-axis [kgm ]
2
γ = ightpath angle [rad] Izz = mass moment of inertia around the z-axis [kgm ]
δa = aileron deection angle [rad] l = length of fuselage [m]
δe = elevator deection angle [rad] L = lift [N] / roll moment [Nm]
δr = rudder deection angle [rad] m = aircraft mass [kg]
δp = throttle setting [0%-100%] M = pitch moment [Nm]
θ = pitch angle [rad] N = yaw moment [Nm]
ρ = atmospheric density [kg/m ]
3 p = roll rate [rad/s]/ atmospheric pressure [Pa]
φ = roll/bank angle [rad] pdyn = dynamic pressure [Pa]
ψ = yaw angle or heading [rad] q = pitch rate [rad/s]
a = speed of sound [m/s] r = yaw rate [rad/s]
b = wing span [m] rxCG = location of the center of gravity on the x-axis [m]
c = mean aerodynamic chord [m] ry CG = location of the center of gravity on the y-axis [m]
C = cross wind force [N] rz CG = location of the center of gravity on the z-axis [m]

1
 CC = cross wind coecient Sref = reference wing area [m ]
2
CD = drag coecient T = thrust [N]/ atmospheric temperature [K]
Cl = roll moment coecient u = velocity component along xB -axis [m/s]
CL = lift coecient v = velocity component along yB -axis [m/s]
Cm = pitch moment coecient V = free stream velocity [m/s]
Cn = yaw moment coecient w = velocity component along zB -axis [m/s]
D = drag [N] xB = x-axis in body system
2
g = gravitational acceleration [m/s ] yB = y-axis in body system
h = altitude [m] zB = z-axis in body system

2 Introduction
Using a simulation program such as Simulink [1] is code. The C code is autogenerated using some help
a useful method to design and analyze complex sys- les custom-made by Mathworks to ensure a success-
tems and its variation over time. Here, the equa- ful interface between Simulink and ARES. For the
tions of motion for a chosen unmanned generic air- earlier mentioned model subsystems to t as sub-
craft are built up as a system of predened discrete models to ARES, inputs and outputs were chosen to
blocks which is a clear and logical way of predicting match the inputs and outputs inserted and outgoing
the behaviour of the aircraft when moving. from ARES.

The aim of this project is to create a simulation The Predator C Avenger [2] was chosen where the

model for an unmanned generic aircraft to study its tail was modied to simplify the calculations and to

ying qualities. The model should then represent the t the description of being a standard aircraft. To

aircraft in a steady state for a chosen altitude and obtain the aerodynamic data needed, a tool called

speed. Disturbances in form of rolling, pitching or SUMO (SUrface MOdeller) [3] was used to create a

yawing are then separately added to investigate the surface model and a surface mesh combined by tak-

reaction of the aircraft due to the changes and hence ing aerodynamic data that are typical for a general

the stability in the longitudinal and lateral directions. aviation airplane.

The following subsections provide more informa-
Aircraft specications and other data could then
tion about the aircraft and the programs used in this
easily be changed so the model can be used to simu-
project.
late other similar aircraft. The Simulink model could
then be used to simulate the motion of an aircraft
and hence its ying qualities. The other applica- 2.1 Aircraft equations of motion
tion is for the parts of it dening the aircraft to act
The equations of motion are derived from Newton's
as submodels to ARES (Aircraft Rigid body Engi-
neering Simulation) which is a simulation tool used
second law F = ma. Those equations represent a
rigid body with six degrees of freedom which can be
by Saab to simulate among other the JAS39 Gripen.
divided into three translation motions and three ro-
Those parts/subsystems connected to ARES can be
tations. In addition to the six equations representing
extracted from the Simulink model and are the aero-
forces and moments, there are three more equations
dynamic model i.e. the one calculating the aerody-
which dene the Euler angles. The dynamic equations
namic forces and moments, the engine model and the
[4] are dened as follows:
inertia model, the other subsystems needed are pro-
vided by ARES instead. The outputs from running
ARES plus those subsystems should then be compara- X − mg sin θ = m(u̇ + qw − rv) (1)

ble to the results from the complete Simulink model.

The Simulink model could then be used as a sim-
Y + mg cos θ sin φ = m(v̇ + ru − pw) (2)
pler simulation model beside ARES. The advantage
of building a graphical model is that it is simpler to
build and review and mainly for those who are unfa- Z + mg cos θ cos φ = m(ẇ + pv − qu) (3)
miliar with programming languages such as C, C++
or Fortran which was used to create ARES. It is also
necessary to be careful when modelling large systems L = I x ṗ − (I y − I z )qr (4)

in Simulink because it is easy to get lost between

the subsystems and the arrows linking them together
M = I y q̇ − (I z − I x )rp (5)
which makes it dicult to nd eventual errors.

The connection between ARES and Simulink is

accomplished by converting the Simulink model to C N = I z ṙ − (I x − I y )pq (6)

2
 where X ,Y and Z are the force components act- If assuming that the engine has no installation an-
ing on the aircraft and consists of aerodynamic forces gle i.e. only acts in the x-axis direction, the force
and thrust and, components can written as:

Cl = Cl0 + Clβ β + Clδa δa + Clδr δr +

X = T − D cos α cos β + C cos α sin β + L sin α (13)

b L
(Clp p + Clr r) = (7)
2V qSref b Y = −D sin β − C cos β (14)

C m = Cm0 + Cmα α + Clδe δe +

Z = −D sin α cos β + C sin α sin β − L cos α (15)

c M
(Cmq q) = (8)
2V qSref c where

Cn = Cn0 + Cnβ β + Cnδa δa + Cnδr δr + CC C C0 + CCβ β + CCδr δr

C= = (16)
qSref qSref
b N
(Cnp p + Cnr r) = (9)
2V qSref b CD C D0 + CDα α + CCδe δe
D= = (17)
Equation (1)-(6) are derived considering the body qSref qSref
frame system i.e the xed axis system of the airplane.
Euler angles θ, φ and ψ in combination with the ve-
dx dy dz c


locity of the aircraft in the xed frame
dt , dt and dt CL C L0 + CLα α + CLδe δe + CLq q 2V
are then necessary to describe the position and orien- L= =
qSref qSref
tation of the aircraft i.e. where the body is relative (18)
to earth and can be formulated as,
Here, a linearized model is assumed which means
" dx # that the aerodynamic derivatives in equation (7)-(9)
dt and (16)-(18) are assumed constant. Due to the lack
dy
dt
=
dz of wind tunnel data, SUMO was used to produce the
dt
aerodynamic derivatives needed to determine among
other the lift coecient. It is then worth mentioning
" #" #
Cθ Cψ Sφ Sθ Cψ − Cφ Sψ Cφ Sθ Cψ + Sφ Sψ u
Cθ Sψ Sφ Sθ Cψ + Cφ Cψ Cφ Sθ Sψ − Sφ Cψ v that the aerodynamic derivatives were assumed to not
−Sθ Sφ Cθ Cφ Cθ w be aected by changes within the interesting range of
speed or altitude. The longitudinal and lateral equa-
θ̇ = q cos φ − r sin φ (10) tions are not coupled to simplify the complexity of
the problem which means for example that the lon-
gitudinal aerodynamic coecients are not aected by
φ̇ = p + q sin φ tan θ + r cos φ tan θ (11) adding an aileron or a rudder deection angle.

ψ̇ = (q sin φ + r cos φ) sec θ (12) 2.2 Predator C Avenger

Where C stands for cos and S stands for sin. The aircraft used is the Predator C Avenger [2] devel-
A result of considering the body axis system is oped by General Atomics Aeronautical Systems. This
that forces acting in the wind frame system have to unmanned aircraft had its rst ight in 2009 and is
be transformed. The transformation process is shown both used in combat and reconnaissance with stealth
as follows: features. It has a turbofan engine and can operate at
high speeds with long endurance. Some specications
   and performance data are presented in Table 1.
cos α cos β − cos α sin β − sin α −D
 −C ⇒ As mentioned in the introduction, the tail was
 sin β cos β 0
modied where the v-tail was replaced by conven-
sin α cos β − sin α sin β cos α −L
tional vertical and horizontal stabilizers. This change
was made to avoid eects such as coupled moments
 
−D cos α cos β + C cos α sin β + L sin α
as the v-tail acts as both a rudder and an elevator.
 −D sin β − C cos β 
Figure 1 shows a three-view sketch of the aircraft com-
−D sin α cos β + C sin α sin β − L cos α
bined with body axes, dierent angles and moments.

3
 Wing span 20.11 [m] 2.3 Simulink
Length 13.4 [m]
Powerplant PW545B (turbofan) As mentioned before, Simulink is used here to model
and analyze the aircraft's dynamic system i.e. its vari-
Thrust 18 [kN]
ation over time. The equations of motion are designed
Max takeo weight 7167 [kg]
using a combination of dierent predened blocks
Max altitude ≈ 15000 [m]
that are connected together with arrows. Those
Max endurance 18 [hr]
blocks are a part of the Simulink library which in-
Max air speed 350 [KTAS]/180[m/s]
cludes the blocks needed to create a system. A system
consists of dierent, linked parts such as inputs, out-
Table 1: Specications for the Predator C [2]
puts, math operations blocks and user-dened blocks
etc. The results from simulating the model can be an-
alyzed for example by exporting the outputs to Mat-
lab's workspace or by adding a scope. An example a
Simulink model can be seen in Figure 2.
The parameters dening the whole system such as
aerodynamic derivatives, aircraft specications, ini-
tial values for velocity and altitude, available thrust
etc have to be stored as global Simulink parameters
even called mpt parameters. Also the inputs and
outputs from the Simulink model that are connected
to the ARES model have to be stored as mpt sig-
nals. Those global Simulink parameters and signals
are then gathered in a m-le which is used as a data
source and is loaded into the model's base workspace
through the model explorer, see Appendix 2.
The Simulink model had to be modelled regard-
ing some requirements from ARES. One demand was
that the model had to be created as a discrete system
with no continuous states and with a xed sample
time which enables the creation of autocode that can
be used in ARES.

Figure 1: Predator C Avenger

Figure 2: Example of a Simulink model

Due to lack of information regarding thrust at dif-

2.4 SUMO (SUrface MOdeller)
ferent altitudes and Mach numbers, the original en-
gine [5] was replaced by another one with a similar SUMO is a graphical tool developed by KTH that
bypass ratio. The new engine is the FJ44-1C (RM- allows a quick and easy creation of surface models.
15) which is installed in Saab's SK60, see Appendix 1. Creating the surface model is done by manually in-
The thrust table for the FJ44-1C was multiplied by 4 serting the design specications of the aircraft such
to ensure that available thrust is higher than required as the length of the fuselage and the wing span, see
thrust as the SK60 has two engines and to compen- Appendix 3. When the surface model is completed, a
sate for the fact that the predator weighs twice as surface or a volume mesh is autogenerated which can
much. be seen in Figure 3 and 4.

4
 The other part is a potential ow solver called
dwfs (Dry W ater F low Solver) [7] from which the
aerodynamic derivatives needed are produced. To be
able to run the dwfsolver, a surface mesh .msh is
needed and also a .cfg document that states dier-
ent initial conditions. Those initial conditions include
among other chosen values of the angle of attack α,
the sideslip angle β, the Mach number and xed val-
ues such as the weight of the aircraft.

SUMO calculates the forces acting on the aircraft

with respect to a reference point dened in the x, y
and z direction. This reference point is assumed to
act as the position of the center of gravity. As the
position of the center of gravity was unknown in this
case, a trial and error process began by guessing the
position of the reference point which had to be located
ahead of the aerodynamic center to guarantee stabil- Figure 3: Surface model by SUMO
ity. Values for the position in the y and z direction
were chosen to be close to zero due to symmetry which
leaves the position on the x axis unknown. The goal
was to vary the reference point until getting a positive
C m0 and a negative C mα which indicates stability. By
testing some dierent positions, it was seen that the
mentioned stability condition could not be fulllled as
it was dicult to get a positive C m0 without getting
a positive C mα too. C m0 is considered as a design
parameter so the design of the aircraft was reviewed
to see if there was any changes that could be made to
improve stability. To examine which parameters that
aect C m0 , the following equation was considered,

Cm0 = ηVH CLαt (ε0 + iw − it ) (19)

lt St
where VH =Sc̄ .
A few attempts were done to achieve C m0 > 0 by Figure 4: Surface mesh by SUMO
increasing the tail area St and the tail incidence angle
it and a possible result was to double the area of the
horizontal stabilizer plus having a negative it = −6. 3 The Simulink model
The position of the reference point was then deter-
mined to be (x, y, z) = (7.0, 0, 0.21) which seems re- The rst step was to identify the dynamic equations
alistic as the length of the fuselage is 13.4m and the of an aircraft which are presented in Subsection 2.1.
engine is mounted on the rear end of the fuselage. The next step was to divide the equations into groups
Now that the position of the center of gravity and model those groups as subsystems in Simulink.
is known, the next step was to run the dwf solver Putting the equations in groups means for example
for some dierent atmosphere densities i.e. altitudes that equation (1)-(3) representing forces are in one
and Mach numbers while having α and β equal to subsystem etc. After creating those subsystems, they
zero. The results consists of a mix of both longi- were connected and the whole model which is called
tudinal and lateral aerodynamic derivatives that are the main model was tested by simulating it during
created due to α, β , angular velocities or deecting time. The results are then sent to Matlab's workspace
the control surfaces (ailerons, elevators & rudder). It to be plotted and analyzed.
was detected that altitude does not aect the aero- The idea of the model is to simulate the motion of
dynamic derivatives while changing the Mach num- the aircraft in a trimmed state i.e equilibrium. For a
ber did change these derivatives. As the aerodynamic chosen altitude and speed, the trim values in the lon-
derivatives were assumed to be constant when chang- gitudinal direction are calculated using a trim system
ing the speed and the altitude, a mean value was cal- called the trim model modelled in Simulink. Those
culated for each of the derivatives. trim values are then inserted to the main model de-

5
scribed earlier. 3.2 The main model
When the part regarding trimming is completed,
This model is designed so that the expression for
disturbances are inserted to investigate stability. Dis-
the rst-order dierential equation is isolated on the
turbances are inserted in form of deecting one of the
right-hand side of the equations so for example equa-
control surfaces. Stability in both longitudinal and
tion (1) looks like following:
lateral directions are examined by looking at the dif-
ferent eigenvalues that are excited during the time the X − mg sin θ
system tries to return to equilibrium. − qw + rv = u̇ (25)
m
The main application of the main model is to be
The x-component of the velocity u is then solved
connected to and act as a submodel to ARES. Only
by integrating using an integrator block. As the
specic subsystems of the model are used during the
model is created assuming no continous states, a unit
interface between Simulink and ARES. Those subsys-
delay block in discrete time is used. A simple way of
tems are the one calculating aerodynamic forces and
integrating in discrete time is done by expressing the
moments combined with an engine model plus an iner-
right-hand expression as following:
tia model. Other parts such as an atmosphere model
is instead provided by ARES. un+1 − un
u̇n+1 = (26)
the
Below follows a more detailed description of 4t
trim model and of the parts creating the main model. =⇒

u̇n+1 4t + un = un+1 (27)

3.1 The trim model
A steady state is assumed here i.e no acceleration or Where 4t is the sample time.
altitude change and the pitch rate q is zero. This The integration process is modelled in Simulink as
model was then designed to calculate the values of shown in Figure 14 in Appendix 5 where the initial
some longitudinal trim parameters where lateral trim value for the velocity component u0 is added in the
parameters such as δa , δr and β are assumed to be unit delay block.
zero, see Appendix 4. The longitudinal parameters As mentioned before, the main model consists
are α, δ e , θ , δ p , u and w. The trim system is mod- of dierent subsystems describing forces, moments,
elled using the following equations: rates, Euler angles and thrust settings etc. The
main subsystem is the one including the aerodynamic
↑: L = mg + Tr sinα (20) forces and moments plus its coecients. The com-
plete model can be seen in Appendix 5.

→: D = Tr cosα (21)
3.2.1 Aerodynamic forces and moments
This subsystem is modelled using equations (1)-(3) to
C m = 0 = C m0 + C mα α + C mδe δe (22) calculate the aerodynamic forces and equations (7)-
(9) to calculate the aerodynamic moments. As ex-
pected, this part only determines aerodynamic forces
and moments i.e no gravity or thrust are included.
where L = (CL0 + CLα α + CLδe δ e )qSref ,
The total forces are calculated in another subsystem
called the total forces model. Beside outputs in form
and D = (CD0 + CDα α + CDδe δ e )qSref .
of aerodynamic forces and moments, dierent coe-
cients and moment coecients in the body system
The equations (20)-(22) are solved for the vari-
are calculated as well. As this subsystem is used
ables α, δ e and T r. To obtain the throttle setting
to connect with ARES, the inputs and outputs from
δp , the required thrust Tr is divided by the available
this subsystems are named so ARES would recognize
thrust TA which is a table with values that depend
them.
on altitude and speed. The ight path angle γ is zero
because there is no altitude change which means that
the trimmed value of α = θ. The velocity components 3.2.2 Aerodynamic rates (angular velocities)
u and w are calculated using: Equations (4)-(6) are used here and unit delay blocks
are applied in the same way as in Appendix 5, Fig-
u = V cosα (23) ure 14 to determine the angular velocities in xB , y B
and z B direction. The aerodynamic moments L, M
and N calculated by the aerodynamic forces and mo-
w = V sinα (24) ments subsystem are inputs to this subsystem. The

6
moments of inertia need to be calculated as well and 3.2.5 Engine
to achieve that following equations [6] are used:
The engine model is designed using the lookup ta-

b2 m(Rx )2 ble (n-D) block which interpolates the existing thrust

Ixx = (28)
table and collects an available thrust value based on
4g
a chosen Mach number and an altitude. The Mach
2 2 number decides which row and the altitude decides
l m(Ry )
Iyy = (29) which column where the appropriate thrust value is
4g
located in the thrust table. The value of the available
thrust is then multiplied by δp to get the required
e2 m(Rz )2 thrust.
Izz = (30)
4g

where Rx , Ry and Rz are non-dimensional radii

of gyration, 3.2.6 Additional subsystems
b+l
and e =
2 .
Here, the total forces are calculated. The total forces
As the values for Rx , Ry and Rz for the Preda-
include thrust, aerodynamic forces and gravity. The
tor are unknown, values that correspond to a simi-
outputs are the velocity components u, v and w .
lar aircraft were used instead. Rx , Ry and Rz were
Those outputs are then used to determine α, V , β ,
calculated for a STOL transport [4] Table B.28 and
γ and h. α and V are calculated using equation (23)
then used to calculate the moments of inertia for the
and (24). The other outputs are obtained using:
Predator.

v
3.2.3 Euler angles β = arcsin √
u2 + v 2 + w2
(36)

To calculate the Euler angles, the rates subsystem is

connected to this one and the rates p, q and r are
inputs used to solve equation (10)-(12). γ =θ−α (37)

3.2.4 Atmosphere model

The International Standard Atmosphere model (ISA) dz
is used as an atmosphere model. It is assumed that h=− = −(−u sin θ + v sin φ cos θ + w cos φ cos θ)
dt
the aircraft will not y above the Tropopause i.e it (38)
ies under an altitude of 11 [km]. This condition
gives:

T = T0 + L(h − h0 ) (31)
4 Results
The results are divided into dierent sections. The
T −g rst section presents the results when simulating the
p = p0 ( ) LR (32)
T0 aircraft in a trimmed state. As mentioned earlier, the
trim model calculates the trimmed values needed as
T −g inputs to simulate the main model. The other sections
ρ = ρ0 ( )( LR −1) (33)
T0 show the results of applying changes to the model i.e.
how the aircraft reacts on changes such as deecting
one of the control surfaces. Longitudinal and lateral
stability are investigated for the dierent changes.
p
a = γRT (34)

pdyn = 0.5ρV 2 (35) 4.1 Trim

where Lr is the lapse rate (−6.5

o C/km)
For some chosen altitudes and speeds, the values
Velocity and altitude control this model and the needed for the aircraft to stay in equilibrium are ob-
outputs are the dynamic pressure pdyn and the speed tained by the trim model. Table 2 shows the results
of sound a. T , p and ρ are only calculated to deter- from the trim model where h and V are controls and
mine pdyn and a. α, δ e , θ, δ p , u, w and CL are outputs.

7
 Chosen trim inputs (1) (2) (3) to lateral disturbances. Dierent disturbances are in-
Mach [−] 0.3 0.4 0.5 serted to the main model to excite the ve modes and
Altitude [m] 3000 3000 3000 the variables aected by the disturbances are plotted
Trim outputs (1) (2) (3) to compare the results with the eigenvalues estimated
α [deg] 3.3 1.28 -0.02 from the longitudinal and the lateral matrix A.
δ e [deg] 1.93 2.41 4.7
θ [deg] 3.3 1.28 -0.02
δ p [−] 43% 69% 77%
u [m/s] 98.4 131.4 164.3 4.3 Longitudinal stability
w [m/s] 5.68 2.92 0.96
C L [−] 0.452 0.255 0.164
Longitudinal stability is dened as the stability
around the lateral axis of the aircraft and is even
Table 2: Trim values used in the main model
called pitching stability. The longitudinal motion is
disturbed from the trimmed state i.e. equilibrium and
As the system is in equilibrium, the expected
is characterized by two oscillatory modes called the
result should be that no changes occur when plot-
short period mode and the Phugoid.
ting outputs over some time. When choosing a new
speed and/or altitude as a new trim value, the system
should then nd equilibrium.

4.3.1 Short period mode

4.2 Dynamic stability
Dynamic stability denes the aircraft's tendency to This mode is important because it indicates how fast
return to its original position after being disturbed the aircraft responds to longitudinal changes. The
without the interference of a control system. Before short period mode is heavily damped with a short
returning to original position, a serie of oscillations period and the aerodynamic derivatives aecting this
occur. If the oscillations are being damped out dur- mode are C mq , C mα and C mα̇ . To be able to detect
ing time then the dynamic stability is described as the short period mode, an elevator doublet, which
positive i.e. the system is stable. Eigenvalues or does not excite the Phugoid, is inserted to the sys-
modes describe the aircraft's unforced response to an tem, see Appendix 6.2.2. The variables α and θ are
applied disturbance and can be determined by solving aected by the elevator doublet and are plotted to
the equation look at the short period. This mode is fast enough
that the velocity and the altitude do not have time to

[λI − A] = 0 (39) change and remain therefore constant. The value of

the short period mode from the longitudinal A ma-
where A is a linearized matrix containing the air- trix for M = 0.3 and h = 3000 was calculated to
craft's stability derivatives. There are two linearized λsp = −1.67 ± 1.625i which gives,
matrices, one containing the longitudinal derivatives
and the other one with the lateral derivatives, see Ap-
t1/2 = 0.415 [s], N1/2 = 0.11 cycles and period =
pendix 6.1, Figure 38.
3.87 [s].
λ can be written as The real part is negative which indicates stability.
Figure 5, shows the elevator doublet added to the

λ = η ± iω (40) system where the short period mode should appear

after the dotted lines marking the forced response.
where η is the real part called the total damping It is hard to observe this mode because of the high
and is used to calculate the time required to halve damping resulting in only a tenth of a cycle before
(stable motion) or double the amplitude (unstable the amplitude is halved. The damping is high mainly
motion), thalf /double = 0.693
|η| , due to the high value of C mq . By decreasing the value
and ω is the imaginary part called the damped of C mq , the oscillations due to the short period ap-
natural frequency of the system and is used to calcu- peared when plotting α, see Appendix 6.2.1. The dif-
2π
late the period of the oscillations, period =
ω . ference between the values of t1/2 from the plot and
The number of cycles to half/double the ampli- from the calculated eigenvalue was about 13% and
ω
tude can be expressed as Nhalf /double = 0.110
|η| . the dierence regarding the periods was about 2% .
There are ve modes, see Appendix 6.1, the short The dierences could depend on the dierences that
period mode and the Phugoid that are excited due to occur when comparing a linearized model i.e. the A
longitudinal disturbances and the roll mode, the spi- matrix and the more detailed Simulink model created
ral mode and the Dutch roll mode that are excited due using nonlinear equations.

8
 about 20% comparing the time to half and about 44%
comparing the period. The conclusion made was the
possibility that C Du and C Lu are not equal to zero.
These derivatives were calculated using the following
approximative method,

M2
CDu = CDM =0 (41)
1 − M2

M2
CLu = CL (42)
1 − M 2 M =0
where CDM =0 and CLM =0 were assumed to have
the same values as CD and CL calculated using equa-
tion (17)-(18) from the Simulink model. The results
can be shown in Table 3, where the derivatives were
Figure 5: Detecting the short period by plotting α at calculated for three Mach numbers.
Mach=0.3
Mach 0.3 0.4 0.5

4.3.2 Phugoid C Du 0.046 0.034 0.0124

C Lu 0.39 0.16 0.054
This eigenvalue is also called the long period mode
and can be recognized by looking at the changes in Table 3: Value of C Du and C Lu for dierent Mach
θ, h and V. Unlike the short period, the Phugoid numbers
is characterized by a lightly damped motion during
a long period of time which gives the pilot enough The values for C Du and C Lu were inserted to the
time to counteract this motion. Adding an elevator Simulink model as well as to the longitudinal A ma-
step is a way of exciting this mode. The aerodynamic trix. Changes in C Du and C Lu had a little impact
derivatives aecting this mode are C D0 , C L0 , C Du on the Simulink model. The modied A matrix gives
and C Lu where C Du and C Lu are assumed to be zero λp = −0.0117 ± 0.0875i resulting in,
in this case. The Phugoid from the longitudinal A t1/2 = 59.2 [s], N1/2 = 0.82 cycles and period =
matrix forM = 0.3 and h = 3000 was calculated to 71.8 [s].
λp = −0.0087 ± 0.057i which gives, It is then seen that changing the values of C Du
t1/2 = 79.6 [s], N1/2 = 0.72 cycles and period = and C Lu results in reducing the dierence between
110 [s]. the two t1/2 to about 3% and improves the dierence
between the periods from 44% to 15%.
Applying the same method at M ach = 0.4 helps
decreasing t1/2 56.8 [s] to 42.5 [s] which is more
from
similar to the value of 41 [s] obtained from plotting θ
and also decreases the period from 111.6 [s] to 89 [s]
compared to 82 [s], also determined from plotting θ .
For the case of M ach = 0.5, it only diers about 6.4%
comparing t1/2 obtained from the Simulink model and
the one obtained from the A matrix and about 9%
comparing the periods. Adding C Lu slightly improves
the dierence between the periods from 9% to 6.6%
while adding C Du results in increasing the dierence
from 6.4% to 33.3%.
The reason for not getting exact values compared
to the values from the Simulink model depends on
dierent uncertainties. One reason is that equations
(41) and (42) are approximative. Another reason
Figure 6: The Phugoid motion is the dienence that can occur when comparing a
Simulink model created by modelling nonlinear equa-
Figure 4.3.2 gives t1/2 = 61 [s] and period = 61 [s]. tions and the A matrix that is obtained from lin-
Comparing t1/2 and the period from Figure 4.3.2 with earizing the dierential equations of motion using the
the calculated eigenvalue shows that the results diers small-disturbance theory.

9
4.4 Lateral-directional stability moving in a tighter and tighter spiral. By inserting a
small roll angle change to the system, the spiral mode
Lateral stability is stability around the longitudinal
can be detected by plotting the roll angle φ. This
axis (roll axis) and directional stability is the stabil-
mode has no imaginary part i.e. no oscillations so
ity around the vertical axis (yaw axis). The lateral-
it converges towards or diverges from the trim value.
directional modes consists of coupled rolling and yaw-
The aerodynamic derivatives aecting this mode are
ing motions. The measured lateral modes from the
C l β , C l r , C nβ and C nr . For M = 0.3 and h = 3000,
Simulink model are also here compared to the ones
the spiral mode was calculated to λspiral = −0.013
obtained from the linearized lateral A matrix.
which gives t1/2 = 53 [s].

4.4.1 Roll mode

Roll mode can be dened as the damping of the roll
motion i.e. the time it takes for the system to reach
steady state roll rate when a disturbance is added.
By adding an aileron step, the roll mode can be seen
when plotting the roll rate over time and the measure
of the system's response can be calculated. This re-
sponse is called the roll mode time constant τroll and
is the time needed to reach 63% of the steady state
roll rate. C lp is the derivative aecting the value of
the roll mode and for M = 0.3 and h = 3000, the
roll mode was calculated to λroll = −2.87. τroll can
1
also be determined by using τroll = −
λroll which gives
τroll = 0.35 [s].

Figure 8: The spiral mode

As seen from Figure 4.4.2, t1/2 is around 70 [s]

which diers about 24% from t1/2 calculated earlier
from the eigenvalue. This dierence is reduced to
about 10% when considering higher speeds. Trying
to calculate the spiral mode approximately, see Table
5, gives a poor result compared to the spiral mode
values obtained from Simulink and ARES. This could
be an explanation to the dierence in the results as
the A matrix is assumed to produce approximative
solutions.

Figure 7: The roll mode

As seen in Figure 7, not only the roll mode appears

4.4.3 Dutch roll mode
but also the Dutch roll mode and τroll is estimated to
This mode is a combination of rolling and yawing os-
0.3 [s]. This results in a dierence of about 14% com-
cillations where the airplane yaws in one direction and
paring τroll from Figure 7 and from the calculated
eigenvalue. The dierence is smaller for higher speeds
rolls in the opposite direction when a β disturbance
occurs, resulting in the aircraft rocking from side to
and is around 8%.
side in combination with a tail-wagging. The Dutch
roll mode is an undesired motion and consists of a
4.4.2 Spiral mode real and an imaginary part. This mode is aected by
all the lateral and directional stability derivatives and
Investigating the spiral mode enables detecting the
tendency of the aircraft to roll towards or from the
for M = 0.3 and h = 3000 the Dutch roll mode was
wings level. A slowly increasing roll angle results
calculated to λspiral = −0.13 ± 1.25i which gives,
in increasing the sideslipping and the aircraft starts t1/2 = 5.3 [s], N1/2 = 1.06 cycles, p = 5.0 [s].

10
 6% comparing the two α angles from Simulink and
ARES. The other trim parameters have all dierences
below 5% so it is the assumed that the values of the
trim parameters match suciently enough.

The next step is to analyze the results of compar-

ing the dierent modes determined using Simulink,
ARES and the approximative solution. From Table 5,
the modes obtained from Simulink and ARES have an
exact match regarding the short period mode, the roll
mode and the spiral mode. Looking at the Phugoid,
the time to half matches exactly and the periods dif-
fers about 10%. The 10% dierence corresponds to
7 seconds which is a short time compared to ∼ 60
Figure 9: The Dutch roll mode seconds i.e. the period of the Phugoid. The Dutch
roll mode diers too with about 7% comparing the

As seen from Figure 4.4.3, t1/2 = 5.0 [s] and p= time to half but the periods match exactly on the

5.0 [s] which gives a dierence of about 5.7% compar- other hand. The 7% dierence corresponds to ∼ 0.4
ing the time to half and no dierence comparing the seconds which also is a short time compared to the

periods. For higher speeds, the results of comparison period of 5 seconds.

are almost the same as for the case of M = 0.3.

4.5 Simulink vs Ares The approximaive solutions dier more compared

to the values obtained from ARES which is expected
The nal step was to extract the submodels that are because as the name indicate, these solutions are ap-
autocoded and connected to ARES. The submodels proximative but should anyway give an indication of
are, as mentioned in the introduction, the aeromodel how the values of the modes are going to be. The
(calculation of aerodynamic forces and moments), the most noticeable dierence regards the spiral mode's
engine model and the inertia model i.e. the sections approximative solution of −0.14 compared to −0.01.
that dene the aircraft, see Appendix 7 in Figure 82. According to [4], the approximative solution for the
Other submodels such as the atmosphere model are spiral mode is considered poor which could explain
instead provided by ARES. The Simulink submod- the large dierence.
els are converted into C code using the earlier men-
tioned, custom-made les that automatically generate
the C code using Matlab. The next step is to merge
that code with ARES and the trim process can start For higher speeds, similar results are obtained

by chosing a trim altitude and a trim Mach number. regarding both the comparison between ARES and

In the same way as when simulating using Simulink, Simulink and between ARES and the approxima-

steps can be added to the control surfaces deection tive solutions. The approximative solution for the

angle and dierent variables aected by the changes Phugoid diers however more for higher speeds and

can be plotted to measure the system's modes. up to 50% for M ach = 0.5.
The idea is to investigate the results obtained
from simulating the complete model through Simulink
compared to simulating using ARES connected to
the chosen subsystems. The comparison between
Simulink and ARES was regarding if the same trim
values were obtained as well as the dierent modes.
The comparison of the results for M ach = 0.3 and M=0.3/h=3000 Simulink ARES

h = 3000 can be seen in Table 4 & 5. Appendix 7 α [deg] 3.3 3.1

shows the same results for M ach = 0.4 and M ach = θ [deg] 3.3 3.1
0.5. Approximative solutions [4] regarding the modes δe [deg] 1.93 1.97
are also included and are calculated for the three δ p [−] 43% 45%
speeds.
Starting with the trim parameters in Table 4, the Table 4: Comparing the trim parameters obtained
largest percentage dierence is calculated to be about from Simulink and from ARES

11
The short period • The surface modeller SUMO and the potential
Simulink ARES Approximation
ow solver dwfs are recommended tools when
−1.67 ± 1.625i −1.67 ± 1.625i −1.67 ± 1.624i lacking aerodynamic data. It is easy to model a

The Phugoid surface model using the aircraft specications

Simulink ARES Approximation data and the mesh is autogenerated. When

−0.012 ± 0.09i −0.012 ± 0.10i −0.015 ± 0.12i the initial conditions are set, the potential ow
solver is easily run. The noticeable obstacle is
The roll mode the lack of an instruction manual.
Simulink ARES Approximation
−3.30 −3.30 −2.80 • If designing the Simulink subsystems regarding
some requirements from ARES, then the sub-
The spiral mode
Simulink ARES Approximation
systems dening the aircraft - i.e. the aerody-
namic model (forces and moments), the engine
−0.01 −0.01 −0.14
model and the inertia model (mass and mo-
The dutch roll mode ments of inertia) - can be converted into au-
Simulink ARES Approximation togenerated C code and then act as submodels
−0.13 ± 1.25i −0.14 ± 1.25i −0.07 ± 1.25i to ARES. The no continuous states requirement
led to the unit delay block being used instead
Table 5: Comparing the dierent modes at Mach=0.3
of the integrator block.

• By obtaining the values for the longitudinal

5 Conclusions trim parameters and inserting them to the main

Simulink model it was shown that the system
managed to nd equilibrium and disturbances
In this project, the dynamics of a modied Predator
were then easily added by inserting a step to
C Avenger were modelled using the graphical simula-
one of the control surfaces deection angle.
tion tool Simulink. The Simulink model is designed
as a steady state system where disturbances in form • The results in form of the ve eigenvalues ob-
of deecting the control surfaces were added to study tained from the Simulink model were compared
the ying qualities of the aircraft. Results in form to the ones obtained from the linearized A ma-
of the ve dierent eigenvalues were compared to the trices and if the dierence between the two
approximative solution determined from the longitu- methods was not large then the Simulink model
dinal and lateral linearized A matrices [4]. The sub- was assumed to work properly. How large the
models dening the aircraft were extracted from the dierences are can be seen in Appendix 6.7.
Simulink model and integrated as submodels to the The largest dierence regards the spiral mode at
simulation program ARES. The results from simu- M=0.3 and is about 24% and could be explained
lating through ARES were compared to the results by the fact that the spiral mode approximation
from simulating the complete Simulink model through is assumed poor. The other dierences were at
Simulink. The aim was to examine the accuracy of the most about 15% and are assumed acceptable.
designed Simulink model. The conclusions obtained
from this project are: • By comparing the results from running the com-
plete Simulink model, the results from simulat-

• Simulink is an easy and logical tool that can be ing through ARES and the approximative so-

used to model the dynamics of an aircraft i.e. lutions, it can be assumed that the Simulink

its motion variation during time using prede- model is suciently accurate as it matches the

ned blocks from the Simulink library and link- results from ARES with a few deviations.

ing them together with arrows to create systems

and subsystems. This means that no program-
ming is needed, simplifying for those who are
6 Future work
unfamiliar with programming languages.
• The Simulink model can be extended by adding
a control system to regulate the behaviour the
• It is also necessary to be careful when construct-
system.
ing large systems in Simulink as it can be dif-
cult and time-consuming to nd potential er- • A pilot model can be designed which for an UAV
rors. is more of designing a joystick control model.

12
7 Acknowledgments
I would like to thank my supervisor Ph.D Andreas Nilsson for all the help and support and my thesis work
would have never been completed without his guidance. I would also like to thank Roger Larsson and Mårten
Staaf at Saab for all their help.
From KTH, I would like to thank Ph.D David Eller for helping me with SUMO and Professor Ulf Ringertz
for taking time to evaluate my thesis work.

8 References
[1] Mathworks, Simulink, 2011-09-05, http://www.mathworks.se/products/simulink/index.html
[2] Wikipedia, General Atomics Avenger, 2011-09-15, http://en.wikipedia.org/wiki/General_Atomics_Avenger
[3] Larosterna, Sumo, 2011-10-01, http://larosterna.com/sumo.html
[4] R. C. Nelson, 1998, Flight stability and automatic control, second edition
[5] Pratt & Whitney Canada, PW545B, 2011-10-15, http://www.pwc.ca/en/engines/pw545b
[6] J. Roskam, 1989, Airplane design, Part V: component weight estimation, second edition

[7] Larosterna, dwfs, 2011-10-01, http://larosterna.com/dwfs.html

13