Budapest University of Technology and Economics

Faculty of Transportation Engineering and Vehicle Engineering

Department of Control for Transportation and Vehicle Systems

Distributed propulsion system based control

of a morphing wing UAV

Author Supervisor
Mocsányi Réka Dóra Takarics Béla PhD.

Budapest, 2020
Summary

Hybrid-electric, unconventional aircraft solutions can possibly be the solutions for the

ambitious emission reduction targets set by regulators, based on society's demands. One

such disruptive solution is a morphing wing cargo UAV, with distributed propulsion. This

paper investigates the aerodynamics, ight dynamics and control of a scaled down tech-

nology demonstrator UAV, built to validate the feasibility of the morphing wing concept.

Several types of analyses are run to gain knowledge on the performance, stability and con-

trol properties of the aircraft. The ight mechanical eects of the distributed propulsion

system are taken into account based on the integral momentum theorem. The increased

ow speed behind propellers increases the local lift forces. Therefore, the distributed

propulsion can be used to control the roll, pitch and yaw motion of the morphing wing

aircraft. The nonlinear 6 degrees of freedom, distributed propulsion aircraft model is

constructed utilizing the stability and control derivatives obtained from the aerodynamic

analysis. Grid and Tensor Product (TP) type linear parameter-varying (LPV) models of

the morphing wing aircraft are generated via Jacobian linearization and TP model trans-

formation. The LPV models capture the parameter varying dynamics arising from the

airspeed, morphing wing and payload weight variations. Classic gain scheduled lateral

and longitudinal baseline controllers are synthesized using the grid-based LPV model of

the aircraft, using the conventional control surfaces. Since the motion of the aircraft can

also be controlled using this specic propulsion system, a second baseline controller is

designed based on the propellers. To obtain the eciency of the propulsion based base-

line controller compared to the classic baseline controller, both are used to navigate the

aircraft along a previously determined path.

Tartalmi kivonat

Hibrid repül®gép kialakítások jelenthetnek megoldást a légi közlekedés üzemanyag fo-

gyasztásának és károsanyag kibocsátásának csökkentésére. Egy elosztott propulziós rend-

szerrel ellátott változtatható szárnyú teherszállító pilóta nélküli repül®gép jó példaként

szolgálhat az ilyen repül®gépek viselkedésének és tulajdonságainak szemléltetésére és vizs-

gálatára. A dolgozatban vizsgált járm¶ egy ilyen repül®gép arányosan kicsinyített reprezen-

tációja, mely a koncepció megvalósíthatóságának tanulmányozása céljából épült. Az elosz-

tott propulziós rendszer hatásának modellezése az impulzus elmélet alapján történik. A

propulzió által indukált áramlási sebesség növekmény a légcsavar mögötti szárnyfelület

részeken felhajtóer® növekedést eredményez. Ez a hatás a repül®gép orsózó, bólintó és

legyez® mozgásának szabályozására is alkalmas, mivel a légcsavarok teljesítménye külön-

külön történik. Különböz® aerodinamikai és repülésmechanikai szimulációk révén ele-

mezhet® a járm¶ hatékonysága, stabilitása és irányíthatósága. Ezen vizsgálatok során

kapott aerodinamikai, stabilitás és kontroll tényez®ket felhasználva jön létre a hat sz-

abadságfokú, nemlineáris, elosztott propulziós rendszer¶ repül®gép modell. A grid alapú

és Tensor Product (TP) típusú lineáris paraméter függ® (LPV) modellek Jacobi lin-

earizáció, illetve TP modell transzformáció révén állnak el®. A létrehozott LPV mod-

ellek magukba foglalják a rendszer dinamikájának paraméter függ®ségét, ezáltal képe-

sek alkalmazkodni a repülési sebesség, a szárny húrhossz és a hasznos teher nagyságá-

nak változásaihoz a repülés során. A grid alapú LPV modellre gain-scheduled típusú

hossz- és keresztirányú baseline szabályozó is készül, mely a klasszikus módon, kor-

mányfelület kitérítésével irányítja a repül®gépet. A légcsavarok elkülöníthet®ségét és ezál-

tal a repül®gép mozgásának szabályozhatóságát kihasználva a klasszikus baseline irányítás

mellett tervezhet® egy a propulzió hatását felhasználó baseline kontroller is. A repül®gép

nemlineáris modelljét egy megadott pályán végigvezetve kiértékelhet® a hajtóm¶vekkel

történ® irányítás hatékonysága a klasszikus szabályozáshoz képest.

Acknowlegments

This work was supported by Hungarian National EFOP-3.6.1-16- 2016-00014 project ti-

tled Investigation and development of the disruptive technologies for e-mobility and their

integration into the engineering education (IDEA-E).

The research leading to these results is part of the FLiPASED project. This project

has received funding from the European Unions Horizon 2020 research and innovation

programme under grant agreement No 815058.

Special thanks to Tamás Baár and Dr. Tamás Luspay for their technical support on

my study.

Budapest, 2020.
Contents

1 Introduction 5

2 Aircraft Aerodynamics 8
2.1 Aerodynamic Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

2.2 Nonlinear Aircraft Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

3 LPV Model of the Aircraft 14

3.1 Grid-based LPV model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

3.2 TP type polytopic LPV model . . . . . . . . . . . . . . . . . . . . . . . . . 15

4 Gain Scheduled Baseline Control 17

4.1 Gain scheduled baseline control architecture and design . . . . . . . . . . . 17

4.2 Path following navigation system . . . . . . . . . . . . . . . . . . . . . . . 20

4.3 Simulation results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

5 Conclusion 26

iv
Chapter 1

Introduction

There are signicant changes aecting the world of aviation these days. As with many

other elds of life today, society's desire for greener, more ecient and immediate, on

demand transportation solutions are presenting challenges in the industry and act as a

non-classical market pull force. Due to societal pressure regulators and stakeholders have

dened ambitious future targets, for example emission reductions in the order of 90% and

75%, in terms of N OX and CO2 respectively.

Based on today's understanding of technology levels, and technology forecast, this

radical improvement can't be achieved using incremental development alone. Aviation

is required to begin its next development phase, and needs novel, disruptive technolog-

ical solutions and out-of-the-box thinking to achieve the radical targets. Electric and

hybrid-electric powered aviation can be an answer, as seeing the trends in other forms

of transportation, electric propulsion has the potential to achieve the target when imple-

mented correctly.

The morphing wing aircraft presented in this study is the subject of an unconventional

solution to the challenges associated with electric and hybrid-electric propulsion systems

[1]. The morphing mechanism is conceptualised as a lightweight solution that is capable

of supporting the air loads, can be adjusted to account for the varying mass of the cargo

UAV and can be used as ecient aps and ailerons. In order to validate the feasibility

of such mechanism, in 2019 a technology demonstrator UAV was designed and built at

the Department of Aeronautics, Naval Architecture and Railway Vehicles of Budapest

University of Technology and Economics. The UAV in addition to validating the mor-

phing concept can be tted with a regular wing, in which case it can be used alongside

other departmental UAVS, such as the micro-gasturbine test UAV [2], for research and

educational purposes. The cargo UAV concept, along with the technology demonstrator

is shown in Figure 1.1.

 1. FEJEZET. INTRODUCTION

Figure 1.1: Cargo UAV (left) and technology demonstrator UAV (right) models

The key parameters of the UAV are the following:

ˆ MTOM: 2kg

ˆ Wingspan: 1.4m

ˆ Wing area: 0.28 − 0.31m2

ˆ Speed range: 32 − 80km/h

The rst contribution of the paper is the aerodynamic modeling and analysis of the

morphing wing aircraft using XFLR-5 software [3]. In addition to the performance anal-

yses, the assessment of the longitudinal, as well as the lateral stability of the aircraft is

concluded. As a rst approach to take the distributed propulsion system of the aircraft

into account, the ow around the wing is adjusted according to the integral momentum

theorem [4]. The ow acceleration caused by the propellers has a positive inuence on

the lift generated by the wing, therefore enhancing the aircraft's ight performance. The

aerodynamic analysis and the distributed propulsion modeling of the morphing wing air-

craft is presented upon which the nonlinear 6 degrees of freedom (DoF) model can be

created.

A common way of control synthesis for aircraft is utilizing gain scheduling and the

linear parameter-varying (LPV) framework [5, 6]. There are three main representations for

LPV systems. These are the grid-based LPV models [7], linear fractional transformation

(LFT) [8] and polytopic [9, 10, 11, 12, 13] models. All these LPV representations have

their advantages and disadvantages, therefore, they complement each other. The present

paper focuses on grid and polytopic based LPV models and the goal is to develop such

6
 1. FEJEZET. INTRODUCTION

LPV models of the morphing wing aircraft. The polytopic model is derived via Tensor

Product (TP) model transformation. TP model transformation is an ecient numerical

method capable of transforming LPV models into TP type polytopic models. The main

underlying principle behind the TP model transformation is the higher order singular

value decomposition (HOSVD) [10]. The grid based LPV model serves as a basis for the

baseline control design and the implementation of a path following algorythm. Lateral

and longitudinal, gain scheduled baseline controllers are derived for the morphing wing

aircraft based on the considerations of [14], and the path following controllers are built

based on [15].

7
Chapter 2

Aircraft Aerodynamics

The dynamic model of the aircraft requires the aerodynamic, stability and control co-

ecients of the vehicle. For this an open-source aerodynamic analysis tool, XFLR-5

was used [3]. The program rst requires the geometry and load distribution of the air-

craft, based on which the dierent analyses can be run. After constructing the models,

performance analysis of the aircraft can be executed to determine general aerodynamic

properties and information on its static stability in steady ow. After achieving statically

stable models, dynamic stability and control analyses are carried out in order to gain the

stability and control derivatives of the vehicle. The obtained coecients - aerodynamic,

stability and control - are then utilized in the dynamic nonlinear 6 DoF model created in

Matlab/Simulink.

2.1 Aerodynamic Analysis

Performance analysis requires the plane surface to be separated to panels, since it is

carried out using the vortex lattice method, which is a potential based panel method.

According to [16], the basic concept of panel methods is that a large number of elementary

quadrilateral panels with one or more types of singularities attached to them are placed on

the actual or mean surface of the geometry in question. A functional variation across the

panel can be specied to determine the singularities, for which the actual value is set by

the corresponding strength parameters. The parameters can be calculated by solving the

tting boundary condition equations. The vortex lattice method (VLM), rst formulated

in the 1930's, is similar to the standard panel methods introduced above. Considerable

dierences listed in [17] are that VLM's formulations ignore the thickness of surfaces

using calculations based on combinations of thin lifting surfaces, boundary conditions are

applied on a mean surface and singularities are not distributed over the entire surface.

It is basically a simple method with a purely numerical approach and the advantage of

computational eciency.

In order to model this morphing wing properties of the aircraft accurately, several

8
 2. FEJEZET. AIRCRAFT AERODYNAMICS

versions representing dierent wing chord length settings and payloads are required. To

achieve a sucient range of models, discrete points are chosen, at which performance,

stability and control analyses are executed. In this case 3 chord settings and 3 payload

weights are chosen resulting in a sum of 9 discrete models. The middle chord length is

at 220 mm from which the length can be modied by 30% in each direction. The airfoils

used for the morphing wing aircraft are presented in Figure 2.1. As for the payload carried

by the vehicle, the weight ranges from 0 kg to the maximum load of 1.5 kg . Additional

payload was included to investigate the eect of the aicraft's weight on each version of the

wing. The parameter values applied in the nine separate models are presented in Table

2.1.

Figure 2.1: Morphing airfoil

Table 2.1: Parameter variance for the discrete models

Varying Discrete parameter values

parameters lower boundary mean upper boundary
Chord c̄ (a) 154 mm (b)220 mm (c) 286 mm
Payload mpayload 0 kg 0.75 kg 1.5 kg

With each version of the aircraft, the properties of interest are calculated with the

airspeed xed at 22 m/s during the performance analyses. These simulations were run at
◦ ◦
dierent angles of attack ranging from −5 to 5 . The results for the 3 unloaded models
◦ ◦
are shown in Figure 2.2. The trim angle of attack is between −0.0437 and 0.0001 and the

corresponding lift coecients are positive. The slopes of the Cm − α curves are negative,

indicating the longitudinal stability of each aircraft.

9
 2. FEJEZET. AIRCRAFT AERODYNAMICS

0.8
c = 154mm
c = 220mm
0.6 c = 285mm

0.4

CL
0.2

−0.2
−4 −2 0 2 4
alpha [deg]

0.1
c = 154mm
c = 220mm
5 · 10−2 c = 285mm

0
Cm

−5 · 10−2

−0.1

−0.15
−4 −2 0 2 4
alpha [deg]

Figure 2.2: CL and Cm vs. α graphs

2.2 Nonlinear Aircraft Model

The dierent congurations of the aircraft require separate dynamic representations to

account for the morphing of the wing, as well as the dierent payloads applied to the

airframe. Similarly to the models created in xr-5, 9 discrete versions with dierent chord

settings and payloads are generated. A nonlinear 6 DoF dynamic model of the morphing

wing aircraft consists of two subsystems: one of them contains the implementation of

traditional 6 DoF ight mechanics model for aircraft and the other calculates forces and

moments that act on the airframe. The ight mechanics model describes rigid body

motion with translational velocities u, v , w and angular rates p, q , r in the body-frame,

while orientation is dened in the inertial referenc frame with the Euler angled φ, θ , ψ .

10
 2. FEJEZET. AIRCRAFT AERODYNAMICS

The angle of attack α and sideslip angle β are the angles between the body and the

wind axes, while the ight path angle γ and course angle χ describe the ight path with

respect to the inertial frame. The previously obtained aerodynamic, stability and control

coecients are applied in the latter element of the model, that includes gravitational,

aerodynamic and propulson eects. However, since this vehicle generates thrust using

a distributed electric propulsion system, the eect of these devices on the aerodynamics

of the wing requires greater attention. Previous studies have shown that a larger ow

speed increased by the propulsion system, especially by a distributed propulsion system,

results in a signicant benet to the lift coecient [18, 19, 20]. To take the eect of

the propellers into account, the lift coecient distribution on the wing obtained via the

XFLR-5 performance analysis was modied to match the increased ow speed generated

by the propellers. To generate a distribution that contains this impact, the wing area is

separated into 5 surfaces. As seen of Figure 2.3. the separate areas are the ones behind

the 4 propellers and the fth is the sum of the remaining surface of the wing. The dynamic

L1 L2 R2 R1

Figure 2.3: Distribution of the wing surface

pressures at the 5 surface can be dierent from each other, which will then directly eect

the local lift force. The integral momentum theorem is used to compute the ow speed

behind each of the propellers individually. According to this theorem, there is a sudden

increase in the speed and pressure at the surface of the propeller [4]. Data on the required

thrust presented in Figure 2.4. is needed to calculate the speed increment.

The thrust force T can be calculated as

T = ρR2 π(V + v)v3 , (2.1)

where ρ is the air density, V is the velocity of the ow, v is the speed increment directly

after the propeller and v3 is the speed increment of the ow where the pressure decreases

to its initial value. As shown in [4], the distant induced velocity v3 is twice in magnitude

compared to the local induced velocity v. Using this fact, the thrust force becomes

T = 2ρR2 π(V · v + v 2 ). (2.2)

11
 2. FEJEZET. AIRCRAFT AERODYNAMICS

6
c = 154mm
c = 286mm
5

Required thrust [N]

4

1
5 10 15 20 25 30 35
Airspeed [m/s]

Figure 2.4: Required thrust for the whole aircraft at 2 dierent chord setting [21]

After rearranging the quadratic equation to its standard form, it becomes

T
v2 + V · v − = 0. (2.3)
2ρR2 π

The speed increment directly after the propeller can therefore be directly obtained and

can be used to calculate the local dynamic pressures behind each propeller. With such

an approach, the local lift forces can be computed. This modeling method allows the

propellers to operate separately, which enables dierences in the magnitude of thrust

generated by each of them. As a result, a higher thrust force on one side of the wing

causes a yawing and rolling motion of the aircraft. At the same time the increased lift

on the corresponding area of the wing generates a momentum around the longitudinal

axis, enhancing the motion caused by the larger thrust force. Additionally, using such a

model allows for safer operation in case of a motor failure as well. This complex propul-

sion model is implemented beside the aerodynamic and gravitational force and moment

calulation.

Additionally, the assessment of lateral stability is achieved using the nonlinear aircraft

model, by the inspection of the dynamic modes of the aircraft. In Figure 2.5. the pole

migration of an example version of morphing wing aircraft can be observed, while the

properties of the dynamic modes of the same version are detailed in Table 2.2. It can be

seen from the pole migration diagram that each dynamic mode, both longitudinal and

lateral, are stable with the exception of the spiral mode at lower airspeed. Table 2.2.

shows that the damping of the spiral mode is a negative number indicating its instability.

12
 2. FEJEZET. AIRCRAFT AERODYNAMICS

20 22

Short Period
20
15

Airspeed [m/s]
18
Im

10
Dutch Roll 16

5
Roll 14
Phugoid Spiral

0 12
−20 −15 −10 −5 0 5
Re

Figure 2.5: Dynamic modes of the aircraft at V = 20 m/s, c = 220 mm, without payload

Table 2.2: Dynamic mode properties at V = 20 m/s, c = 220 mm without payload

Mode Damping Frequency Time constant

[rad/s] [ s]
Phugoid mode 0.188 0.576 9.23
Short period mode 0.496 14.2 0.142
Roll subsidence mode 1 17.6 0.0568
Dutch roll mode 0.1 4.83 2.06
Spiral mode −1 0.0188 −53.2

13
Chapter 3

LPV Model of the Aircraft

The LPV framework is well suited to capture the parameter dependent dynamics of the

aircraft. In the case of the morphing wing aircraft, in addition to the airspeed, the wing

morphing and payload weight have signicant eect on the dynamics. The aim is to derive

an LPV model that capture all these eects. The formal denition of the LPV model

([5, 6]) is given as

ẋ = A(ρ(t))x + B(ρ(t))u
(3.1)
y = C(ρ(t))x + D(ρ(t))u

where A : P → Rx , B : P → Rx , C : P → Rx , C : P → Rx are continuous matrix

function. x : R → R denotes the state vector, u : R → R the input vector, y : R → R

denotes the output vector and ρ : R → P is a time varying scheduling signal, where P is
ρ
a compact subset of R .

3.1 Grid-based LPV model

The grid-based LPV model consists of linear time-invariant (LTI) models (Ak , Bk , Ck , Dk ) =
(A(ρk ), B(ρk ), C(ρk ), D(ρk )).
These LTI models are obtained by evaluating the LPV
Ngrid
model over a nite grid of the scheduling parameter values ρk 1 = Pgrid ∈ P . The grid-
based LPV model is derived from the nonlinear aircraft model via Jacobian linearization.

For this the scheduling parameter is dened rst in the following way

 
V (t)
ρ(t) =  c̄(t) (3.2)
 

mpayload (t)

The rst step of Jacobian linearization is trimming the aircraft. This is done for straight
   
and level ight condition at the range of airspeed V = 12.5, 22 m/s, chord c̄ = 154, 286
 
mm and payload mpayload = 0, 1.5 kg at 39, 3 and 3 equidistant points, respectively.

14
 3. FEJEZET. LPV MODEL OF THE AIRCRAFT

The next step is to obtain the LTI model sets over these scheduling grid to form the grid-

based LPV model. The pole migration of the LPV model of the morphing wing aircraft

is shown as the function of the airspeed with c̄ = 220 mm and mpayload = 0 kg in Figure

2.5.

3.2 TP type polytopic LPV model

The TP type polytopic LPV model is obtained from the grid based LPV model via

TP model transformation ([10]). The highest singular values in each dimension of the

scheduling parameter after the HOSVD are

 
6447.3
     
 455.6  6437.7 6417.6
 
σV =  28.6  , σc̄ =  485.9  , σmpayload =  706.5 
     
 
 2.62  310.6 302.4
 
0.25

It can be noted that the singular values of σV become rather small beyond the third

singular value. Since the TP model transformation can set a trade-o between accuracy

and complexity ([10]), keeping the rst 3 singular values in this dimension is expected to

result in a TP type LPV model with low complexity but sucient accuracy. Such huge

decrease in the singular values cannot be observed in the morphing chord and payload

dimension which means that the number of vertices of the polytopic model can not be

reduced in these dimension without a signicant loss of accuracy. The CNO type weighting

functions of the TP type LPV model are shown in Figure 3.1. As a result, the grid based

LPV model of the aircraft consisting of 39 × 3 × 3 LTI models was transformed to a TP

type convex polytopic LPV model with 3 × 3 × 3 vertex systems. Such polytopic model

can signicantly reduce the computational cost of the control synthesis that is usually

related with grid-based LPV models at the expense of more conservative results.

15
 3. FEJEZET. LPV MODEL OF THE AIRCRAFT

1
0.8
0.6
0.4
0.2
0
13 14 15 16 17 18 19 20 21 22
Airspeed, V [m/s]

1
0.8
0.6
0.4
0.2
0
160 170 180 190 200 210 220 230 240 250 260 270 280
Chrod, c̄ [mm]

1
0.8
0.6
0.4
0.2
0
0 0.2 0.4 0.6 0.8 1 1.2 1.4
Payload, mpayload [kg]

Figure 3.1: CNO type weighting functions of the TP model

16
Chapter 4

Gain Scheduled Baseline Control

The previously described grid-based LPV model is now available for control design. As

introduced earlier, this aircraft model enables a part of lateral dynamics, namely rolling

motion, to be controlled by the asymmetric operation of the distributed propulsion sys-

tem. In this section baseline control design is completed using a classical approach and

a distributed porpulsion system based approach as well. In addition to a gain-scheduled

baseline controller, a path following algorythm is implemented in each control system to

promote a higher level comparison of the baseline control systems. To achieve autonomous

path following of a UAV, the pilot delegates aircraft control to the ight control system.

A ight path has to be generated on which the reference commands can be based. In ad-

dition to the dynamic model, second order actuator models and nonlinear engine models

are added to the aircraft representation in order to realistically reect the behaviour of

both devices. These include limitations of the input magnitude provided by all control

equipment.

4.1 Gain scheduled baseline control architecture and

design

As mentioned earlier, the morphing mechanism and the applied payload can only be

adjusted on the ground. This means that during ight the dynamics of the aircraft

varies solely according to airspeed. Flight velocity therefore becomes the only required

scheduling parameter in the gain scheduled control during ight simulations. Two versions

of the baseline control are created. The rst type of control, referred to as classic control,

takes inputs from the traditional control surfaces, the aileron, elevator and rudder to

control rolling, pitching and yawing motion of the aircraft respectively. The baseline

controller was taken form [15] and applied to the morhing wing aircraft. In this version

the engines are only in charge of airspeed control. These control systems are built for the

aircraft with mean parameters, namely 220 mm chord length and 0.75 kg payload.

17
 4. FEJEZET. GAIN SCHEDULED BASELINE CONTROL

Controllers are designed based the assumption of decoupled dynamics, meaning that

a separate longitudinal and lateral control system is built. Classic control generates

thrust commands (δt,l1 , δt,l2 , δt,r1 , δt,r2 ) for airspeed control and elevator (δe ) commands to

regulate ight altitude via the longitudinal control. Simultaneously, rolling and yawing

motion is regulated by rudder (δr ) and aileron (δa ) commands, which are produced by

the lateral control system. The feedback loop of the longitudinal control is the pitch-

attitude controller, which tracks pitch attitude, while reducing wind disturbances and

achieves short period damping using the pitch rate as an input. As for the lateral control

the feedback loop is a roll-attitude controller that tracks roll attitude and achieves roll

dalmping using the roll rate as an input. The rudder in only used to maintain 0 sideslip.

Both controllers are scheduled by the ight velocity (V ).

Regarding the distributed propulsion based control compared to the classic version,

the main dierence is that the role of the aileron is taken over by the the two outer engines,

which are therefore partly in charge of the lateral control as well as controlling the velocity

of the aircraft. Lateral control in this case includes rolling and yawing motion control

via asymmetric thrust (δt,l2 , δt,r2 ) and rudder commands. Longitudinal control remains

the same as the one used to design the classical control system. The resulting thrust

commands can be dened as

δt,l1 = δt,r1 = δt
δt,l2 = δt + 0.5δa (4.1)

δt,r2 = δt − 0.5δa .

Figure 4.1 presents the frequency response from both the left aileron δa,l and the

outer left engine δt,l2 to the roll angle φ. The dierence in gain shows that the aileron

has a greater eect on rolling motion. This indicates that regarding the current aircraft

conguration the eciency of the propulsion system based control will be less compared

to that of the classic control system.

A scheduled, structured control design approach was selected based on [22] to address

the baseline control problem. Being a gain scheduled controller, the controller parameters

stored in a vector K depend on the scheduling parameters in vector ρ, which in this

case consists of the airspeed. To reduce complexity, K(ρ) is conned to polynomial basis

functions of the parameters in ρ. Additionally, since airspeed is the only scheduling

parameter, the parameter space is only one dimensional and vector K(ρ) simplies as

K(ρ) = K0 + K1 ρ + K2 ρ2 (4.2)

for each separate controller implemented in the model. Vector K is computed for each

separate controller in the system using a MATLAB software tool, SYSTUNE.

The responses of the LPV model for the given roll reference command using both

18
 4. FEJEZET. GAIN SCHEDULED BASELINE CONTROL

100

Magnitude (dB) 50

−50

−100
−90

−135
Phase (deg)

−180

−225

−270
10−3 10−2 10−1 100 101 102 103
Frequency (rad/s)

Figure 4.1: Bode diagram δa,L ( ) and δt,L2 ( ) to φ

15
Reference
Classic control
10 DPS control

5
phi [deg]

−5

−10

−15
0 5 10 15 20 25 30 35 40 45 50
time

Figure 4.2: Responses for roll reference command

types of control systems is presented in Figure 4.2. As predicted the control performance

is higher of the classic method, however the propulsion based control does not have a

large amount of lost ground to make up.

19
 4. FEJEZET. GAIN SCHEDULED BASELINE CONTROL

4.2 Path following navigation system

To promote a higher level comparison of the baseline control system using classic control

surfaces to the one using the distributed propulsion system, a path following algorythm is

implemented in both control sequences. Path following is a simple navigation capability

to be expected of autonomous vehicles. The problem depicted in Figure 4.3 is as follows:

given a previously determined path, usually straight line or circular orbit paths, with

the vehicles initial location and heading angle. The aim of the algorythm is to compute

the commanded heading angle ψd as the vehicle moves along its course. An important
variable, the cross-track error d can be described as the smallest distance between the
vehicles position and the path. Considering a straight line path along which a set of

waypoints can be dened, the line-of-sight angle θ is formed by the previous and coming

waypoints Wi and Wi+1 . The aircraft is considered to have reached its target waypoint

if it ies within the globe dened by the waypoint and a chosen radius around the given

waypoint. It is practival to choose the same radius for each waypoint within the ight

path. Instead of directly moving toward the next waypoint the implemented algorythm

uses a vitrual target point s for the vehicle to follow. The performance of path following

is higly dependent on the path parameter δ. The general objective is to calculate the

commanded heading angle such that d→0 and |ψ − θ| → 0 as t → ∞.

Figure 4.3: (a) Straight line path following (b) circular orbit following

The previous problem assumes the altitude to be constant. However, it is advanta-

geous to achieve a ight control system that enables 3 dimensional ight path following.

This means that in addition to the reference course angle command χ, reference ight

altitude command Href is computed by the path following algorythm as well. The addi-

tional reference airspeed command ensures that the ight velocity does not disrupt the

path following capabilities of the controller. Both longitudinal and lateral control require

multiple loops to achieve the required altitude and course angle. In the longitudinal con-

trol the outer feedback loop controls the altitude by providing reference pitch attitude

input for the inner loop. The outer feedback loop of the lateral control regulates the

20
 4. FEJEZET. GAIN SCHEDULED BASELINE CONTROL

course angle by computing reference roll attitude input. Control design is carried out rst

for the inner, then for the outer loops. Autonomous ight is achieved by the outer loops

of the control system. This type of control architecture is shown in Figure 4.4 and 4.5.

Figure 4.4: Classic path following control architecture [15]

Figure 4.5: Distributed propulsion system based path following control architecture

The navigation system is a more advanced ight control system that produces com-

mand signals to navigate the aircraft along a dened ight path. The command signals

are generated by a state machine, which takes the longitudinal (xa ) and lateral position

(ya ) of the aircraft as inputs. In this study the ight pattern made up of several straight

21
 4. FEJEZET. GAIN SCHEDULED BASELINE CONTROL

line paths and is dened by 7 waypoints connecting these straight lines. The radius of

the globes around each waypoint was chosen to be 20 m. The path was selected such that

the capabilities of the control system can be examined during the ight simulation. The

coordinates of the waypoints and their corresponding airspeed values are listed in Table

4.1.

Table 4.1: Waypoint coordinates and corresponding airspeed

Waypoint xe [m] ye [m] ze [m] V [m/s]

Wp1 0 0 100 initial speed
Wp2 250 500 0.142 initial speed
Wp3 500 1000 0.0568 20
Wp4 1500 500 2.06 20
Wp5 500 −1500 −53.2 22
Wp6 −500 −1000 9.23 20
Wp7 −250 −500 9.23 initial speed

4.3 Simulation results

Figure 4.6 and 4.7 show that both the classic and the propulsion based control system are

able to suciently follow a predetermined ight path. A slight dierence in the eciency

of path following is observable, as the distributed propulsion based control requires more

time to adjust after each turn. The ight trajectory diers the least in the variations

of altitude, which is due to tha fact that longitudinal control remains the same within

each control architecture. According to the path following control simulations it takes

between 350 − 400 s for the aircraft to return within the 20 m radius of its initial position.
The longer adjustment time can be seen in Figure 4.8. The largest dissimilarity can be

observed in the control inputs provided by the control systems. An obvious change occurs

in the throttle command shown in Figure 4.9, where the additional thrust generated by

the lateral control system is added to that from the speed tracking control system. A

signicant growth can be observed in Figure 4.10 in the rudder response to the sideslip

produced by the lateral control system in case of the distributed propulsion based control.

This is due to the engines having a greater eect on the yaw rate of the aircraft. However

this deection increment does not exceed the limitations of the rudder, thus it is not

considered a critical disadvantage of the examined control system.

22
 4. FEJEZET. GAIN SCHEDULED BASELINE CONTROL

2,000 2,000

1,000 1,000
North [m]

North [m]
0 0

−1,000 −1,000
−2,000 −1,000 0 1,000 −2,000 −1,000 0 1,000
East [m] East [m]

(a) Classic control (b) DPS control

Figure 4.6: Path following

130 130
Altitude [m]

Altitude [m]

120 120

110 110

100 100

−1,000 0 1,000 2,000 −1,000 0 1,000 2,000

North [m] North [m]

(a) Classic control (b) DPS control

Figure 4.7: Altitude tracking

23
 4. FEJEZET. GAIN SCHEDULED BASELINE CONTROL

Reference
22
Classic control
DPS based control
21

20
u [m/s]

19

18

17

16
0 50 100 150 200 250 300 350 400 450 500
time

Figure 4.8: Airspeed measurements compared to command signal

25
Classic control
24 DPS contol

23

22
t,l2

21
δ

20

19

18

17
0 50 100 150 200 250 300 350 400 450 500
time

Figure 4.9: Throttle command

24
 4. FEJEZET. GAIN SCHEDULED BASELINE CONTROL

3
Classic control
DPS based control
2

1
[deg]

0
δ r

−1

−2

0 20 40 60 80 100 120 140 160 180 200

time

Figure 4.10: Rudder command

25
Chapter 5

Conclusion

A 6 DoF nonlinear, distributed propulsion model of the morphing wing aircraft was

constructed. The aerodynamic stability and control derivatives were obtained from the

XFLR-5 software. The eect of the distributed propulsion system was taken into account

using the the integral momentum theorem. A grid based LPV model was obtained by

Jacobian linearization. The scheduling parameter vector includes the airspeed, morphing

wing chord and the payload mass. TP type polytopic model of the morphing wing air-

craft is obtained from the grid based model via TP model transformation. A signicant

drop in the number of the vertex in the airspeed dimension was achieved. Finally, gain

scheduled inner loops of the baseline controllers and path following outer control loops

were developed for a classic and a disrtibuted propulsion based control system applied the

grid based LPV model of the aircraft. The closed loop lateral and longitudinal dynam-

ics of the aircraft have suciently high bandwidth and the lightly damped short period

mode has increased damping to reduce oscillatory motion. Comparing the classic and

the distributed propulsion based verions of the baseline and the path following controllers

it can be concluded that the classic method is a more eective approach, however the

propulsion based method does not remain far behind. Thus aircraft design that aims

to utilize the propulsion system as a means of control instead of the traditional control

surfaces is a solid option for future studies on the morhing wing concept. Future steps of

the research include investigating fault detection algorithms for the distributed propulsion

system, optimal propulsion bending and advanced Kalman-lter development.

26
Bibliography

[1] D. Sziroczak, I. Jankovics, I. Gal, and D. Rohacs, Conceptual design of small aircraft

with hybrid-electric propulsion systems, Energy, vol. 204, p. 117937, 2020.

[2] K. Beneda and B. Sipula, Design of an ultra-high bypass ratio fan stage for a research

turbojet engine, Acta Avionica, vol. 21, 2019.

[3] A. Deperrois, Xr-5, open source vlm software, 2019.

[4] T. Gausz, Szarnyprol, szarny es legcsavar vizsgalata, Repulogepek €’s Hajok Tan-
szek kiadvanya, 1995.
[5] J. S. Shamma, Analysis and design of gain scheduled control systems. PhD thesis,

Massachusetts Institute of Technology, Cambridge, 1988.

[6] G. Becker, Quadratic stability and performance of linear parameter dependent sys-
tems. PhD thesis, University of California, Berkeley, 1993.
[7] F. Wu, Control of Linear Parameter Varying Systems. PhD thesis, Univ. California,

Berkeley, 1995.

[8] A. Packard, Gain scheduling via linear fractional transformations, Systems & Con-
trol Letters, vol. 22, pp. 7992, Feb. 1994.
[9] P. Apkarian, P. Gahinet, and G. Becker, Self-scheduled h∞ control of linear

parameter-varying systems: a design example, Automatica, vol. 31, pp. 12511261,

Sept. 1995.

[10] P. Baranyi, Y. Yam, and P. Varlaki, Tensor Product Model Transformation in Poly-
topic Model-Based Control. CRC Press, 2013.
[11] L. Kov€cs and G. Eigner, Tensor product model transformation-based parallel

distributed control of tumor growth, Acta Polytechnica Hungarica, vol. 15, Mar.

2018.

[12] V. C. da Silva Campos, L. M. S. Vianna, and M. F. Braga, A tensor product

model transformation approach to the discretization of uncertain linear systems,

Acta Polytechnica Hungarica, vol. 15, Mar. 2018.

27
 BIBLIOGRAPHY

[13] Y. Kan, Z. He, and J. Zhao, Tensor product model-based control design with relaxed

stability conditions for perching maneuvers, Acta Polytechnica Hungarica, vol. 15,
Mar. 2018.

[14] R. W. Beard and T. W. McLain, Small unmanned aircraft: theory and practice.
Princeton University Press, 2012.

[15] T. Luspay, D. Ossmann, M. Wuestenhagen, D. Teubl, T. Baár, M. Pusch, T. M. Kier,

S. Waitman, A. Ianelli, A. Marcos, B. Vanek, and M. H. Lowenberg, Flight control

design for a highly exible utter demonstrator, in AIAA Scitech 2019 Forum,
American Institute of Aeronautics and Astronautics, Jan. 2019.

[16] K. Towell, Aerodynamics for engineers  sixth edition j. j. bertin and r. m. cum-

mings pearson education limited, edinburgh gate, harlow, essex, CM20 2je, UK.

2013. 832pp. illustrated. ¿64.99. ISBN 978-0-273-79327-4., The Aeronautical Jour-

nal, vol. 118, pp. 455456, Apr. 2014.
[17] R. M. Cummings, W. H. Mason, S. A. Morton, and D. R. McDaniel, Applied Com-
putational Aerodynamics. Cambridge University Press, Apr. 2015.
[18] K. A. Deere, S. Viken, M. Carter, J. K. Viken, M. Wiese, and N. Farr, Computa-

tional analysis of powered lift augmentation for the LEAPTech distributed electric

propulsion wing, 2017.

[19] K. A. Deere, J. K. Viken, S. Viken, M. B. Carter, M. Wiese, and N. Farr, Computa-

tional analysis of a wing designed for the x-57 distributed electric propulsion aircraft,

in 35th AIAA Applied Aerodynamics Conference, American Institute of Aeronautics

and Astronautics, June 2017.

[20] M. Selig, Modeling propeller aerodynamics and slipstream eects on small UAVs in

realtime, in AIAA Atmospheric Flight Mechanics Conference, American Institute of

Aeronautics and Astronautics, Aug. 2010.

[21] B. Horvay, Aerodinamika, repulesmechanika, sulypont, 2019.

[22] P. Sujit, S. Saripalli, and J. Sousa, An evaluation of UAV path following algorithms,

in 2013 European Control Conference (ECC), IEEE, July 2013.

28