Creating an Autonomous Octocopter Royal Institute of Technology Bachelor thesis

Creating an Autonomous Octocopter Royal Institute of Technology Bachelor thesis
Royal Institute of Technology
Bachelor thesis
Creating an Autonomous
Octocopter
Author:
Alexander Grima
Jonathan Fagerström
Supervisor:
Arne Karlsson
School of Engineering Sciences
June 18, 2014
Abstract
This report covers the basic approach to creating a new multirotor
design. All the way from creating a concept, through design requirements and finally evaluation. Both control algorithms and system
design is covered. The main problem solved is that of the danger of
naked propellers used on most multirotor drones and how to make
them safer and easier to fly in populated areas.
Contents
Introduction
1.1 Background . . . . . . . . . . . . . . . . . . . . . . .
1.2 Problem . . . . . . . . . . . . . . . . . . . . . . . . .
1.3 Purpose . . . . . . . . . . . . . . . . . . . . . . . . .
Concept
2.1 Case study . . . . . . .
2.2 Concept development
2.3 Concept choise . . . .
2.4 Dimensioning . . . . . .
Powertrain
3.1 Motor . .
3.2 Propeller
3.3 ESC . . . .
3.4 Battery .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Control
4.1 Control algorithms
4.2 Thrust steering . . .
4.3 Micro-processor . . .
4.4 Sensors . . . . . . . . .
.
.
.
.
.
.
.
.
2
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
4
4
4
4
.
.
.
.
6
6
6
8
10
.
.
.
.
11
11
12
14
15
.
.
.
.
17
17
17
17
18
.
.
.
.
20
20
21
25
25
Conclusions
6.1 Summary . . . . . . . . . . . . . . . . . . . . . . . . .
6.2 Discussion . . . . . . . . . . . . . . . . . . . . . . . .
6.3 Further studies . . . . . . . . . . . . . . . . . . . .
26
26
26
27
Acknowledgements
28
Division of labour
29
Results
5.1 Control algorithms . .
5.2 Propeller performance
5.3 Duct performance . . .
5.4 Final model . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
A
31
B
32
C
34
3
Introduction
1.1
Background
The current small multirotors all have similar designs. They all use
naked rotors each mounted on beams extending out from a center
construction. This is worrying the public as they do not seem safe
to use in urban areas. Indeed naked rotors rotating at 8000 rpm can
indeed do a lot of damage to a human. As drones become more and
more common reports of crashes are rising. In Australia the BBC
reported of a woman being injured by a drone, see Figure 1-1.
1.2
Problem
The problem faced in this project is how can these systems be made
safer without impacting performance?
1.3
Purpose
By creating a safe design for multirotors, drones will be able to fly
in urban areas without the danger of damaging anything or anyone
in the event of a system failure. Also by creating intelligent control
algorithms the use of multirotors can be made more accessible.
4
Figure 1-1: Drone crashes in to athlete in Australia. [2]
5
Concept
2.1
Case study
To start designing an improved product one must know what one is
trying to improve upon. By studying what is currently on the market
and analysing the advantages and disadvantages of these products a
picture of possible improvements is created. Following is the train of
thought in the concept evolution process.
2.2
Concept development
The first concept produced was of the same style as those currently
present on the market. The control systems and batteries are mounted
on the center construction with ESC’s strapped to the extruding
beams that support the motors. As is visible in Figure 2-1 the rotors are naked and can easily inflict damage.
6
Figure 2-1: First concept of octocopter.
As Alex took a CAD course he produced a new concept. This concept uses the idea of moving the mass of the batteries and controllers
out from the center so as to increase the systems inertia. Theory is
that this would increase stability in the platform and slow down the
effect of gusts and turbulence. However this does not address the
danger of naked rotor blades as can be seen in Figure 2-2.
Figure 2-2: Alex’s CAD course assignment.
7
From Alex’s concept a new idea emerged as the performance gains
of ducts shrouding the propellers was brought to attention. The ducts
enhance propeller efficiency by reducing blade tip losses and containing thrust wash. Despite the increase in weight hopefully the increase
in efficiency counteracts the added weight and so the rotors can be
protected without a loss in performance. As well as this, by integrating the shrouds into the structure a very solid platform can be
achieved, see Figure 2-3.
Figure 2-3: First concept of octocopter with ducted fans.
2.3
Concept choise
The concept that is chosen to evaluate utilises the idea of shrouded
propellers. The research on shrouds has mostly been with most of
the parameters constant (e.g. diameter to height ratio) and the possibility to evaluate the impact of all the different parameters is close
to impossible. So the design was mainly improvised so as to test the
concept. Similarly to aircraft winglets, a shroud will not always increase propeller performance. With Yilmas and Erdems report [4] as
a starting point the idea used is:
• Large inlet to increase inflow volume.
• Constriction at propeller location to leave as little gap as possible between propeller-tip and shroud wall.
8
• Expanding diffuser to increase pressure difference over propeller
disk and pressure difference over trailing duct wall.
The final concept was designed with this in mind as well as adding
structural integrity with the shrouds. The final concept can be seen
in Figure 2-4
Figure 2-4: The concept chosen.
9
2.4
Dimensioning
As the ducts are structurally bearing they have to withstand forces
from the motors and the additional payload. On top of this the ducts
may not deform to an extent that entails the propeller touching the
wall and resulting in catastrophic failure. Finite element analysis (Ansys) has been run on the 3D modells to ensure structural integrity
with loads 20% greater than maximum theoretical thrust generated.
In Figure 2-5 an example of the deformation as analysed in Solid Edge.
Figure 2-5: Early FEM analysis in Solid Edge.
The difficulty faced here is keeping the weight under a level that
allows the craft to lift. The material chosen to use for the prototyping
is a plastic styrene that is easy to mill. This is however brittle and
so the duct walls have a minimum thickness to allow for the manufacturing. In the final design the wall thickness varies slightly over
the hight of the duct between 1 mm – 8 mm and one full duct weighs
in at 450 grams. To test the concept in a full-scale prototype this is
deemed acceptable.
10
Powertrain
A powertrain was chosen so that the aircraft would achieve sufficient
lift and payload capacity for a given flight time.
3.1
Motor
A brushless permanent-magnet synchronous electric motor (PMSM)
was chosen for their power to weight ratio.
Brushless electric motors are powered by a DC electric source via
an ESC (electronic speed controller), which produces a three phase
AC electric signal to drive the motor. To decide what motor to use
several where compared and the required current to drive the propeller
of choice calculated using Eqn. 3-1.
I=
T
τ
(Eqn. 3-1)
I is the current drawn when the propeller requires a torque of
T to achieve the thrust level and τ is the motors torque constant.
This analysis was short as common consensus pointed towards Tigermotors and by analysing a few of their motors within our budget a
choice could be made.
The motor used was a T-motor MT2814, picked for its capacity
and good performance per cost ratio.
11
3.2
Propeller
To transform the rotational power of the motor to vertical lift a propeller is ideal. In this project the assumption that the airspeed over
the propeller disc is close to zero is made. This assumption is made
because the ducts will force the only movement of air over the disc to
be perpendicular to it and in calm conditions vertical drafts will have
small speeds. Along with this we assume vertical speeds of the multirotor will be small and zero when hovering. With airspeeds close to
or zero only static thrust is analysed, making the analysis far easier.
The basic way of calculating thrust generated by a propeller based
on knowledge of the propeller diameter, pitch and aerofoil geometry
is momentum theory [3]. However the propeller manufacturer provide
propeller data [1] and therefor momentum theory was only used to
calculate thrust power in Eqn. 3-5. In the data, thrust and power
coefficients are available, meaning they do not need to be analytically
obtained, see appendix.
The thrust and power calculations are done according to APC
propellers [1]. These however are in imperial units so a conversion
factor has to be applied to get thrust, F , and power, P , in SI units.
F = Ct · (ρ · n2 · D4 )
(Eqn. 3-2)
P = Cp · (ρ · n3 · D5 )
(Eqn. 3-3)
Where Ct is the non-dimensional thrust coefficient, Cp is the equivalent power coefficient, ρ is the air density, n is RPM and D is the
diameter of the propeller.
Generally a propeller is dimensioned according to the flight environment i.e. relative airspeed (freestream velocity). As previously
stated, for this application the airspeed is assumed close to zero. Thus,
only the static thrust is deemed necessary to analyse. Calculated
static thrust using APC data [1] in equation Eqn. 3-2 can be seen in
Figure 3-1. One can note that as well as the thrust increasing with
RPM so does the required power which is calculated with the equation
Eqn. 3-3.
12
P
(Eqn. 3-4)
ω
Torque is calculated in Eqn. 3-4 where T is torque, ω is the rotational velocity and P the power. As current drawn is proportional
to torque more amperage is being draw from the batteries at higher
thrust levels, decreasing flight time. To maximise flight time and lift
the propellers thrust to power ratio for the chosen system are compared. The propeller with the highest ratio was, not surprisingly,
is the propeller designed for "multirotors". However "slow fly" propellers are not far behind.
T =
Thrust and Power curve for 12x4.5MR prop.
Thrust [N]
Shaft Power [W]
Thrust Power [W]
500
[N]
[W]
10
0
2000
3000
4000
5000
RPM
6000
7000
0
8000
Figure 3-1: Calculated static thrust, shaft power and thrust power
vs rpm for the 12 inch propeller.
13
As well as thrust to power ratio the propeller manufacturer advises
certain RPM limits on it’s propellers depending on the design. With
the chosen motor our analysis dictates that a high RPM is required
to utilise the whole motor. Using this limitation "slow fly" propellers
were out of the question leaving only "multirotor" propellers. To
achieve sufficient thrust a 12 inch propeller was necessary.
To get efficiency of the propeller one can look at the ratio of induced power (thrust power), the power resulting in thrust force and
shaft power, the power produced by the motor. In momentum theory
one can derive Eqn. 3-5.
F 3/2
Pi = √
2ρA
(Eqn. 3-5)
Where A is disk area. The ratio between thrust power and shaft
power is propeller efficiency ηprop
ηprop =
F Vi
Pi
=
= 0.47.
Pshaf t
Pshaf t
(Eqn. 3-6)
Which is quite normal for smaller propellers of this type.
3.3
ESC
To power the motor one must convert the DC supplied by the battery
to AC. In this context, AC, alternating current, does not imply a
sinusoidal waveform, but rather a bi-directional current with a square
waveform. Additional sensors and electronics control the ESC output
amplitude (i.e momentum) and frequency (i.e. rotor speed). The ESC
must be able to deliver enough current to the motor. Pulse-width
modulation (PWM) is used to control the ESC.
The ESC used is a T-motor T30A.
14
3.4
Battery
Batteries are crucial to the aircrafts flight time and lift. A bigger
capacity battery leads to a longer flight time but gives more mass to
lift. The battery also has to be able to deliver the amount of current
necessary at any given time. To get maximum energy density while
still being able to discharge at required rate, LiPos were chosen. These
batteries are built with a number of cells increasing discharge current
capability. However these cells cannot be discharged passed a certain
voltage as they will break down. As we are using multiple batteries
(four) with four cells each we have a total of sixteen cells. These have
to be constantly monitored to assure no one cell drops below 3.3V.
There were no products in the market capable of monitoring so many
cells and thus a circuit to achieve this was created, see Figure 3-2.
Figure 3-2: Circuit to monitor cell voltage in LiPo batteries.
The basic idea of the battery monitor circuit is to compare the cell
voltage of each cell and only send through the lowest voltage to the
micro-processor to evaluate when the aircraft only has enough energy
left to fly it back to the starting position.
With the propeller calculations in section 3.2 and the equation
Eqn. 3-7 one can calculate the flight time at different thrust levels.
15
In Eqn. 3-7 t is the flighttime, C is the battery capacity and I is the
amperage usage. Most of the time this aircraft will be at hover and
therefore an amperage value at hover is used.
t=
C
Ihov
(Eqn. 3-7)
In Appendix Figure B-1 one can see the flight time depending on
number of batteries and battery capacity at hover. The same can be
seen in Appendix Figure B-2 though with maximum thrust.
16
Control
4.1
Control algorithms
In creating our own control algorithms it had to be decided what they
should be capable of. It was our intention from the conception of
this project that the control algorithms be intelligent. Using basic
mechanical reasoning algorithms using positional data instead of attitude data where created. The step response of these algorithms are
in section 5.1.
4.2
Thrust steering
The control algorithms generate a required attitude change and to
achieve that this change must be translated into changes in thrust
levels for the different motors. Using a large matrix the change in
attitude are mapped into changes in rpm for each motor.
4.3
Micro-processor
The micro-processor has to be able to read sensor data and process
it alongside executing the control laws and have enough analog input/output pins. To match these requirements the Arduino Due was
chosen. It has an Atmel SAM3X8E ARM Cortex-M3 32-bit CPU
17
running at 84MHz, 4 serial ports, 54 digital input/output pins 12 of
which can be used as PWM outputs and 12 analog inputs. For the
sensor interpretation multiple libraries from the manufacturer have
been used. This however only generates the raw data, to use it a new
library has been written that interprets the raw data.
Figure 4-1: The Arduino Due micro controller board.
4.4
Sensors
To get reliable movement data exact angular-acceleration data and
acceleration data is needed. To achieve this it was decided to use the
Sparkfun 9DOF MPU-9150. It is a nine degrees of freedom sensor
which contains a 3-axis gyroscope, a 3-axis accelerometer and a 3-axis
magnetometer. This sensor also has a built in motion processing unit
not used in this project. It communicates with the Arduino through
a serial i2C-bus.
18
Figure 4-2: The Sparkfun 9DOF MPU-9150.
Furthermore an ultrasonic ranging module HC-SR04 was needed
as landing the vehicle requires more precise distance measurements
than the MPU-9150 could ever produce. It has an accuracy of 3mm
at distances up to 4m. This communicates with the Arduino via two
digital pins, one TRIG and one ECHO. The distance is measured
simply by triggering a pulse and then measuring the time taken to
receive the echo. This is then divided by the speed of sound.
Figure 4-3: HC-SR04 ultrasonic ranging module.
19
Results
5.1
Control algorithms
To evaluate the control algorithms a step response is used. This means
giving an input of one and measuring the time the system takes to
reach this point. In the lower figure of Figure 5-1 the vertical step
response is shown. It takes the system about 5 seconds to reach one
meter.
Figure 5-1: Step response in z, vertical.
20
Again in the lower figure of Figure 5-2 a horizontal step response
can be seen. Here it takes about 10 seconds to move one meter to the
side or forward.
Figure 5-2: Step response in x and y, horizontal.
These results are very acceptable for this project as no optimisation has been performed.
5.2
Propeller performance
The main area of interest with the propellers is the static thrust performance as the vehicle will mostly be in hover. As only the propellers
static thrust is evaluated it is not terribly difficult to measure as only a
scale is needed for the evaluation. The setup for evaluating propellers
can be seen in Figure 5-3.
21
Figure 5-3: The setup used to evaluate propeller performance.
This rig was designed and built by Alex using Meccano. In Figure
5-4 it is more clearly visible how the testing rig works. By calculating
the moment on the hinge O through measurement of the force R and
lengths A and B the thrust Ft is solved for in Eqn. 5-1
Ft = R ×
22
B
.
A
(Eqn. 5-1)
Figure 5-4: A sketch of the testing rig.
Power used is measured using a watt meter and the RPM with a
tachometer. An example of this data is compiled in Appendix Table
C.1 and compared to the theoretical static thrust calculation in Figure
5-5 and Figure 5-6.
The reason behind the systematically lower thrust in the theoretical calculations is suspected to be due to the provided motor values
use a voltage of 14.8. This voltage is at the batteries lowest allowable
and therefore a worst case value. In testing the battery had a voltage
varying between 16.5 and 15 volts. The efficiency of the motor should
increase with a higher voltage as less current is required to generate
enough power and therefore result in more thrust for a certain power
consumption.
23
Comparison between theory and real world measurements 8 inch propeller
7
6
Thrust [N]
5
4
3
2
1
Theoretical
Measured
0
0
20
40
60
80
100
120
140
Power [W]
Figure 5-5: Comparison between theoretical and measured static
thrust on 8 inch propeller.
Comparison between theory and real world measurements 12 inch propeller
18
16
14
Thrust [N]
12
10
8
6
4
2
Theoretical
Measured
0
0
50
100
150
200
Power [W]
250
300
350
400
Figure 5-6: Comparison between theoretical and measured static
thrust on 12 inch propeller.
24
5.3
Duct performance
Using the same setup as above the small scale duct is evaluated and
the results are compared to the unshrouded propeller. The results can
be seen in Figure 5-7.
20
6
18
4
16
2
14
%
Trust [N]
Efficiency gain with duct
8
Thrust with duct [N]
Thrust without duct [N]
Efficiency gain with duct [%]
0
0
20
40
60
80
100
120
12
140
Power [W]
Figure 5-7: Measured efficiency gain with 8 inch duct.
The anticipated efficiency gain was 10%, which is exceeded. Believing these results are indicative with the gain that would be received
with a 12 inch duct, a full scale model was designed with tweaks made
to areas with minor dimensioning flaws.
5.4
Final model
Unfortunately the full scale model has not yet been manufactured as
KTH’s prototype lab has been busy the weeks leading up to end of
term. The plan is to carry on work during the summer.
25
Conclusions
6.1
Summary
The idea behind shrouded propellers seems to work in a non invasive
way i.e. the added weight is compensated by the increase in static
thrust. However we have not managed to complete a full working
model to prove this. If the downsized model is representative the
complete model should meet the requirements. In other words it is
possible to make multirotors safer by using shrouded propellers without loosing performance.
When it comes to the control algorithms the simulations run in
Matlab indicate they are relatively responsive and do what was intended. Again however, not having the full scale model means the
implementation of the algorithms has not been tested in the field.
We purposely did not go into details explaining the design of the
ducts or the control algorithms as we are currently investigating the
possibility of applying for patents in those areas.
6.2
Discussion
The greatest difficulty experienced has been the manufacturing of the
designed system. Since the parts used are not off the shelf products
they need to be prototyped. Doing this with the difficult geometry
26
and weight restrictions posed a greater issue than anticipated. In
talks with the prototype lab it seemed promising, but when it came
to manufacturing the parts problems where met. However these flaws
have been adressed and the model is yet again under production.
In theory the control algorithms are easy to implement but in
reality quite difficult. To get accurate positional data is not easy
when moving in small areas without walls. This has given us quite a
headache, the solution is to refine optiflow-sensoring or heavy filtering
or to move more than a meter at a time.
What surprised us the most in this project was the shear volume
of work required to move a theoretical idea and system design into a
real model. The work required to transform a 3D CAD design into
something that can actually be printed or manufactured was very time
consuming. As was the programming of micro-controllers and sensors,
along with soldering, creating circuits and securing financial support
for the project.
6.3
Further studies
We intend to continue work on the system during the summer. There
is a lot of room for research regarding the design of ducted fans and
shrouded propellers. Something we are currently interested of looking
into for a masters thesis.
27
Acknowledgements
First, we would like to thank our supervisor Arne Karlsson for his
understanding and enthusiasm towards this project. His suggestions
on where to search for information were very helpful and time saving
as were his comments on this rapport.
Second, thanks to Monika Norrby for allowing us to occupy an
area at the composite lab for our testing.
Finally, we must thank KTH Innovation for their tips and help in
manufacturing parts at KTH’s prototype lab.
28
Division of labour
Both collaborators have worked together and taken part in all areas of
the project, however the work distribution has not always been even.
Jonathan has taken a larger responsibility regarding propeller calculations. Alexander has worked more with the CAD-modelling and
programming of control-algorithms as well as a python interface for
micro-controller communication on the Windows operating system.
29
Bibliography
[1] APC.
Apc propeller data.
http://www.apcprop.com/v/
downloads/PERFILES_WEB/datalist.asp, 02 2014.
[2] BBC. Australian triathlete injured after drone crash. http://
www.bbc.com/news/technology-26921504, 05 2014.
[3] John Seddon and Simon Newman. Basic Helicopter Aerodynamics.
Wiley, 3rd edition, 2011.
[4] Serdar Yilmaz and Duygu Erdem. Effects of Duct Shape on a
Ducted Propeller Performance. In AIAA Paper 2013-0803, 2013.
30
Appendix A
APC data [1]
Appendix Figure A-1: A part of propeller data from APC.
31
Appendix B
Appendix Figure B-1: Flight time in hover with different battery
configurations.
32
Appendix Figure B-2: Flight time in max thrust with different battery configurations.
33
Appendix C
ESC input
10
15
20
25
30
35
40
45
50
55
60
65
70
75
80
85
90
RPM
3410
4100
4730
5240
5690
6080
6440
6780
7070
7365
7710
8180
8610
9040
9490
9990
10410
Power [W]
10,8
14,7
19,4
24
28,7
33,3
38,1
43,3
48
53
59,5
68,69
79,4
90,7
103,4
120
132,5
Current [A]
0,68
0,93
1,23
1,52
1,82
2,13
2,44
2,76
3,08
3,41
3,83
4,45
5,16
5,9
6,77
7,9
8,75
Weight [g]
68
109
142
176
218
248
281
314
342
372
411
470
523
589
650
734
791
Force(R) [N]
0,667
1,070
1,394
1,728
2,140
2,435
2,759
3,083
3,358
3,653
4,036
4,615
5,135
5,783
6,383
7,207
7,767
Thrust(T) [N]
0,592
0,950
1,237
1,534
1,900
2,161
2,449
2,737
2,981
3,242
3,582
4,097
4,559
5,134
5,666
6,398
6,895
Appendix Table C.1: Example of data set for 8 × 45M R.
34
Was this manual useful for you? yes no
Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Download PDF

advertisement