Professional Documents
Culture Documents
Validation of Models For Small Scale Electric Propulsion Systems
Validation of Models For Small Scale Electric Propulsion Systems
Validation of Models For Small Scale Electric Propulsion Systems
net/publication/230688310
CITATIONS READS
34 744
3 authors:
Petter Krus
Linköping University
298 PUBLICATIONS 1,679 CITATIONS
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
Research on Hydromechanical Transmissions and Hybrid Motion systems, RHYTHM View project
All content following this page was uploaded by David Lundström on 04 January 2016.
Original Publication:
Lundström, D., Amadori, K., and Krus, P., “Validation of Models for Small Scale Electric
Propulsion Systems”, 48th AIAA Aerospace Sciences Meeting and Exhibit, 4 - 7 Jan. 2010,
Orlando, FL, USA.
DOI: 10.2514/6.2010-483
At Linköping University work has been carried out on automated design and
manufacturing of Micro Air Vehicles (MAV). A dedicated design optimization
framework has been developed. Initial experience has shown that choosing the right
propulsion system has major relevance for the overall performance of the aircraft. The
correctness of the models used to describe each component of the propulsion system is
therefore a matter of great importance. With this knowledge efforts have been made to
validate the propulsion system models. Using a specifically designed test rig a number of
different motors and motor controllers have been tested. The original motor model has
shown discrepancies compared to test data and improvements to the model are
suggested. Furthermore, motor specifications provided by the manufacturers have
proven to be unreliable. Motor controller characteristics have been shown to be complex
and difficult to model.
Nomenclature
BLDC Brushless Direct Current motor
COTS Commercial Off The Shelf
DC Direct Current
EMF Electro-Motoric Force
ESC Electronic Speed Control
I0 Motor zero load current
Kv RPM proportionality constant
n Motor revolution velocity (RPM)
PWM Pulse Width Modulation
Rm Motor internal resistance
Period time in PWM signal
Duty cycle length in PWM signal
Efficiency
I. Introduction
*
PhD Student, Department of Management and Engineering, David.Lundstrom@liu.se, AIAA Student Member.
†
PhD Student, Department of Management and Engineering, Kristian.Amadori@liu.se, AIAA Student Member.
‡
Professor, Department of Management and Engineering, Petter.Krus@ liu.se.
2
Finally, the optimal design is manufactured using a 3D-printer. This is an attractive method for MAV
manufacturing that allows complete freedom as regards airframe shape. Figure 2 is an example of a MAV that
was automatically designed/manufactured using the described framework. The pictured MAV has been used for
flight testing to validate the method4.
Figure 2. MAV used for flight testing and design framework validation
The principle behind the described framework has been shown to work well and as a next step the models
implemented in the computations need to be validated and, where necessary, improved. When working with
MAV optimization, it has become evident that the propulsion system has a significant impact on performance.
There seem to be more to gain on optimizing the propulsion system, than fine tuning the optimal aerodynamics
layout. The mathematical models that have been used for propulsion system modeling are fairly straight forward
and in common used. There is, however, little information on how accurate these models are. It is also
interesting to study approximately how trustworthy component specifications from different manufacturers are.
The database of propulsion system components, used in the design framework, is based solely on manufacturer
specifications. The focus of this paper is to validate the models used to simulate the electric propulsion system.
paper. More elaborate battery models exist5 which take temperature and dynamic effects into account but would
complicate the optimization process tremendously to little gain. The motor and motor controllers are more
important and are paid most attention in this paper.
A. Motor
The motor characteristics are described with the classical motor constants Kv, I0, and Rm. A model of an
electric motor, using these constants, is shown in Figure 4. The electric motor has its RPM (n) proportional to
motor EMF. Kv is the RPM proportionality constant. Losses in the motor are characterized by its internal
resistance Rm and no load current I0.
Rm
Im
+
I
Uemf
n U emf K v (rpm)
Um I0 M P in I m U m (W)
P out I U emf (W)
This is a widely used motor model that was originally developed for traditional brushed DC motors, but
today also is used in brushless DC motors (BLDC). For instance, commercial electric propulsion system analysis
software, such as Motocalc6 and Electricalc7, uses this model. From Figure 4 some basic equations are derived.
The output power and RPM of the motors, can be expressed as functions of motor input voltage and current
according to the following equations.
n K v (U m I m Rm ) (2)
Pout ( I m I 0 )(U m I m Rm )
m (4)
Pin UmIm
Figure 5 illustrates the behavior of the motor model. With a constant supply voltage, the model says that the
RPM should drop linearly with load and that the power output increases following a second order polynomial
curve (equation (1) and (2) above). These curves can quite easily be tested and will be the focus for the practical
experiments.
4
90 25000
n = -Rm Kv I + Kv U
80
70 20000
n (rpm)
60
Pout (w)
15000
50
40
10000
30
20 5000
10 2
Pout = -RmI + (U+I0Rm )I - I0 U
0 0
0 5 10 15 20
I (A)
Figure 5. General characteristics of motor model
B. Motor Controller
The motor controller, or as it is more popularly called, electronic speed control (ESC), controls the input
power to the motor using a PWM signal. By varying the pulse length (or duty cycle length ) the effective
mean voltage to the motor is varied, thereby controlling the power.
T
Uin
t
T
The losses in an ESC are highly dependent on the duty cycle length. The duty cycle length is most easily
described as an electronic throttle. “Throttle” is therefore in layman’s terms used as a synonym for the ESC’s
duty cycle. Power is controlled by varying the duty cycle length between 0% and 100% of the total period time.
At 100% duty cycle, the efficiency of the ESC is close to 100%, but when the duty cycle is reduced, so is
efficiency. An example of what a brushless motor ESC PWM signal looks like in reality is shown in Figure 7.
The signal in Figure 7 was recorded using a 3-channel oscilloscope, and illustrates the output voltage on each
motor cable of the ESC. The left image shows the ESC at full power with the duty cycle set to 100%. The middle
picture shows the ESC running at a duty cycle length of 50% and the last image shows the ESC with minimum
duty cycle (approx. 5%).
Modeling losses in the motor’s ESC are complex. Not much detail is given by any motor controller
manufacturers. The losses are dependent on several factors. The most obvious are resistive losses in the
electronics when the controller is running at 100% duty cycle. These are generally stated by the manufacturer.
One known behavior of motor controllers is that their efficiency is dependent on duty cycle length, being highest
at 100% duty cycle and less at lower duty cycle lengths. This behavior is not specified by motor controller
manufacturers and a model properly describing this behavior has been sought but not been found. The losses at
duty cycles less then 100% can be divided into two types: i) increased losses in the electronics due to the
additional switching in FETs and ii) the PWM signal to the motor causes the motor to run with lower efficiency
than when it is fed with a smooth continuous voltage.
Validation of Models for… 5
III. Experiment
In order to validate the models of the propulsion system a test rig was built (Figure 9). The test rig is built
around a set of carefully arranged strain gauges (Figure 8). A 10 mm diameter aluminum tube of 0.1 mm wall
thickness is used for torque strain measurement. This tube is connected in series with a load cell for thrust
measurement. The strain gauges are connected to a signal conditioner/amplifier which in turn is logged using a
24-bit analog-to-digital converter connected to a laptop.
Thrust strain
gauges
The electric parameters of the motor are measured using a Medusa Research Power Analyzer Pro8. For correct
current readings also at part throttle, large capacitors are attached to the power analyzer’s output. Table 1
summarizes the different parameters that are measured in the rig and with what resolution each parameter is
measured. The rig can be used to measure the performance of each component in the propulsion system and is
prepared to work in a wind tunnel for dynamic propeller measurements.
o
Temperature C 0-200 1
Testing procedures and data analysis for the motor and motor controller are described in the following two
paragraphs.
A. Motor Testing
The motors of interest for MAVs are brushless DC motors and come from the aeromodeling industry. These
motors are preferred to the older type of brushed DC motors because of their higher efficiency and much higher
power density. Both brushed and brushless motors are often described using the same motor model and motor
constants. For brushless motors, however, it is less clear how these motor constants are defined. It is also not
evident that the motor model is as valid for brushless motors as it is for brushed motors, partly because the motor
controller is required to run the motor and consequently affects its characteristics. In order to study the validity
of the motor model, and to get an estimation of the reliability of manufacturers’ motor constants, 7 outrunner
motors, ranging from 4g to 55g, have been tested in the test bench. Table 2 list all motors tested.
6
A test on a motor is conducted by running a series of propellers of varying size while for each propeller varying
the input voltage incrementally in steps from the minimum voltage required by the ESC to either the maximum
allowed voltage for the ESC or until the maximum current allowed on the motor is reached. During this testing
the ESC is always set to 100% duty cycle. Each voltage step is timed to about 15s. The data is then averaged for
each step and summarized into one table where each logged variable is expressed as a function of U. An example
is shown in Figure 11.
8 0,04 0,60
V/A/Krpm
Nm
Eta
60,0 0,50
6 0,03
0,40
40,0
4 0,02 0,30
Pin
Put 0,20
2 0,01 20,0
n
0,10
Eta
0 0 0,0 0,00
20 40 60 80 100 120 140 160 180 200 220 4,000 5,000 6,000 7,000 8,000 9,000 10,000 11,000 12,000 13,000 14,000
s U (V)
Figure 11. Example data: a) Raw data over time b) Averaged data as function of voltage
By slowly increasing voltage in fine steps of 15s the motor has a chance to stabilize to its operating temperature
for each state. In a previous publication3 it was shown that temperature effects can affect motor efficiency by
several percent. Advancing the voltage too quickly would result in efficiencies being logged at higher values
than is realistic for continuous operation. It is the continuous operation properties that are of interest for these
studies.
The processed data has been compared with predictions using the motor model. The power output of a motor
should, accordingly to the theoretical model (Equation 1), follow the shape of the functions illustrated in Figure
12.
Validation of Models for… 7
Motor max
ESC max voltage current
No load
current ESC
min U
Figure 12. (a) General shape of power output function (b) Relevant area
The RPM (equation 2) and efficiency (equation 3) of a motor should correlate to U and I accordingly to Figure
13.
When testing a motor, the entire range of U and I should ideally be mapped to identify the shape of the surfaces
in Figures 12-13. A quick way of mapping some of the functions is to test the motor with a single propeller while
varying the input voltage from min to max. This roughly corresponds to mapping performance along a curved
diagonal cut through the surfaces in Figure 12-13.
It should be mentioned that each controller had its timing programmed to work as efficiently as possible with
high pole outrunner motors. All tested motors are of either the 12 or 14 pole type.
An example of ESC efficiency measurement is shown in Figure 15. The measurement is carried out while a
propeller is driven with the throttle fixed at 100% and as the input voltage is increased. The voltage is then fixed
and a second run is carried out but this time with the throttle varied from 0 to 100% in steps of 8-10%. Figure
15a shows the efficiency of the two runs plotted against propeller RPM. Figure 15b shows the extracted
controller efficiency.
80 1
0,9
70
0,8
60
0,7
50 0,6
Eta
40 0,5
%
0,4
30
0,3
20
0,2
Eta Variable U
10 0,1
Eta Variable Duty Cycle
0 0
2000 3000 4000 5000 6000 7000 8000 9000 10000 0 10 20 30 40 50 60 70 80 90 100
Figure 15. (a) Efficiency runs 1 and 2 (b) ESC efficiency mapped against duty cycle
One small problem is that when running the first test with the throttle at 100% not all controllers work with
voltages low enough to map the motor’s efficiency down to the same low RPM as in the second step where the
throttle is varied. At small duty cycles the efficiency is not of any particular importance, but in order to get some
values the motor model can be used to extend the efficiency curve for the lower voltages.
The interesting question with controller characteristics is to investigate what parameters affect the
controller’s efficiency. Can a model of an ESC be extracted from test data? To examine the properties of ESC
efficiency each controller was tested on a baseline motor driving the same propeller. A baseline controller was
then tested on several motors. Two controllers were also tested in several operating conditions on the same
motor.
Validation of Models for… 9
IV. Results
25000
16V RPM
14V RPM
12V RPM
20000 10V RPM
8V RPM
6V RPM
15000
10000
5000
0
0 1 2 3 4 5 6 7 I (A) 8
90
16V P out
80 14V P out
12V P out
10V P out
70
8V P out
6V P out
60
50
40
30
20
10
0
0 1 2 3 4 5 6 7 I (A) 8
90 90
tested modeled
80 80
70 70
60 60
50 50
40 40
30 30
16V
14V
20 12V 20
10V
10 8V 10
6V
0 0
0 1 2 3 4 I (A)5 6 7 8 0 1 2 3 4 I (A)5 6 7 8
Interestingly, the behavior of the model, in terms of the shape of the curves, seems to be fairly close to reality,
but there are some unacceptable mismatches between the predicted values and the real values. It was unclear if
this was due to incorrectly defined motor constants or the model failing to describe the exact curve shapes. As a
next step an optimization was carried out on the motor constants in order to minimize the difference between
model and measurements. The optimizer was set to minimize the square of the residuals between measurement
data, and model data, for all points on the RPM and Pout curves. The results are given in Figure 19.
20000 80
16V RPM 16V P out
18000 14V RPM 14V P out
70
12V RPM 12V P out
16000 10V RPM 10V P out
8V RPM 60 8V P out
14000 6V RPM 6V P out
50
12000
10000 40
8000
30
6000
20
4000
10
2000
0 0
0 1 2 3 4 5 6 7 8 0 1 2 3 4 I (A) 5 6 7 8
I (A)
90 90
tested modeled
80 80
70 70
60 60
50 50
40 40
30 30 16V
16V
14V 14V
20 12V 20 12V
10V 10V
10 8V 10 8V
6V 6V
0 0
0 1 2 3 4 I (A) 5 6 7 8 0 1 2 3 4 5 6 7 8
I (A)
Figure 19. Model and measurements after motor constants optimization
In the two top graphs the measurements and model predictions are combined. The model predictions are shown
with dotted lines. In the two lower graphs, representing efficiency, the measurements are shown to the left and
predictions to the right. Some interesting notes: The mismatch between model and measurements is much lower
than with the original motor constants but some characteristics stand out, however. The RPM curves of the
measurements all seem close to linear but their slope is not the same for all voltages as in the model. The output
power curves are close to correct shape but they have a little more curvature (bend downwards) close to
maximum current. The biggest and most severe mismatch is observed on the efficiency curves. The
Validation of Models for… 11
measurements show a maximum efficiency of about 78% for all voltage levels. The model has a spread of 68-
80%. The difference between original and new motor constants is shown in Table 4 below
Can the model be improved? The results indicate that the motor constants perhaps should not be constant, but
functions of U and I. To test this, a new optimization was carried out but this time motor constants were
optimized for each voltage level. The motor constants were then plotted against voltage (Figure 20).
Kv (U) Rm (U) Io (U)
1250 700 0,45
600 0,4
1200 0,35
500
0,3
1150 400
mOhm
0,25
Kv
A
300 0,2
1100
0,15
200
1050 0,1
100
0,05
1000 0 0
0 5 10 15 20 0 5 10 15 20 0 5 10 15 20
U(V) U(V) U(V)
It seemed the model could be improved if motor constants were related to voltage, and the relationship appears
to be fairly linear. The no load current I0 is definitely not a constant which can easily be checked by running the
motor at no load with varied input voltage. Figure 21 shows this for the Mfly motor. Not exactly the same values
as in Figure 20 but the relationship is perfectly linear. This is not surprising since I0 includes the effects of
friction on the motor shaft.
0,50
0,45
I = 0,0159U + 0,2073
0,40
0,35
0,30
I(A)
0,25
0,20
0,15
0,10 Measured I0
0,05 Linear curve fit
0,00
4 6 8 10 12 14 16 18
U(V)
To test if the model could be improved, the equations (1) to (3) were rewritten so that I0 and Rm were expressed
as linear functions of U. The motor parameters (now 5) were then optimized. The results are shown in Figure 22
20000 80
16V RPM 16V P out
18000 14V RPM 14V P out
70
12V RPM 12V P out
16000 10V RPM 10V P out
8V RPM 60 8V P out
14000 6V RPM 6V P out
50
12000
10000 40
8000
30
6000
20
4000
10
2000
0 0
0 1 2 3 4 5 6 7 8 0 1 2 3 4 5 6 7 8
I (A) I (A)
12
90 90
tested modeled
80 80
70 70
60 60
50 50
40 40
30 30
16V 16V
14V 14V
20 12V 20 12V
10V 10V
10 8V 10 8V
6V 6V
0 0
0 1 2 3 4 5 6 7 8 0 1 2 3 4 5 6 7 8
The improved equations show much better correlation with test data, especially for the important efficiency
curves. Still more improvements could likely be made if the motor constants’ relationship to current were
investigated. For instance, the power output curves still have a little more curvature at their upper end than the
model suggests. This is likely due to increased motor temperature at higher currents causing increased resistive
losses. In reality, Rm is not constant but dependent on temperature. Slightly simplified, ignoring transient effects,
cooling factors etc, this dependency could be expressed by having Rm as a function of I.
70,0 70,0
6000
60,0 60,0
5000
50,0 50,0
4000
40,0 40,0
3000
30,0 30,0
2000
20,0 20,0
Measured data Measured data Measured data
1000 Corrected parameters 10,0 Corrected parameters 10,0 Corrected parameters
Original parameters Original parameters Original parameters
0 0,0 0,0
2 3 4 5 U(V) 6 7 8 9 2 3 4 5 U(V) 6 7 8 9 2 3 4 5 U(V) 6 7 8 9
Himax HC2808-0980:
10000 100,0 90,0
n (rpm) Pout (W) (%)
9000 90,0 80,0
8000 80,0
70,0
7000 70,0
60,0
6000 60,0
50,0
5000 50,0
40,0
4000 40,0
30,0
3000 30,0
20,0
2000 Measured data 20,0 Measured data Measured data
Corrected parameters Corrected parameters 10,0 Corrected parameters
1000 10,0
Original parameters Original parameters Original parameters
0 0,0 0,0
2 4 6 U(V) 8 10 12 14 2 4 6 U(V) 8 10 12 14 2 4 6 U(V) 8 10 12 14
Hyperion HP-Z2205-38:
16000 80,0 80,0
n (rpm) Pout (W) (%)
14000 70,0 70,0
7000 50
20
6000
40
5000 15
30
4000
10
3000
20
2000
Measured data 5 Measured data Measured data
10
1000 Corrected parameters Corrected parameters Corrected parameters
Original parameters Original parameters Original parameters
0 0 0
2 3 4 5 U(V) 6 7 8 9 2 3 4 5 U(V) 6 7 8 9 2 3 4 5 U(V) 6 7 8 9
Hyperion Z1705-14:
18000 60 80
n (rpm) Pout (W) (%)
16000 70
50
14000
60
12000 40
50
10000
30 40
8000
30
6000 20
20
4000
Measureed data 10 Measured data Measured data
2000 10
Corrected parameters Corrected parameters Corrected parameters
Original parameters Original parameters Original parameters
0 0 0
2 4 6 U(V) 8 10 12 2 4 6 U(V) 8 10 12 2 4 6 U(V) 8 10 12
AEOrc A05-2900:
35000 25 80
n (rpm) Pout (W) (%)
30000 70
20
60
25000
50
15
20000
40
15000
10
30
10000
20
Measured data 5 Measured data Measured data
5000 Corrected parameters Corrected parameters 10 Corrected parameters
Original parameters Original parameters Original parameters
0 0 0
2 4 6 8 U(V) 10 12 14 16 2 4 6 8 U(V) 10 12 14 16 2 4 6 8 U(V) 10 12 14 16
Some conclusions that can be drawn from the tests: All motors performed with lower efficiencies than could be
predicted with the manufacturer’s motor constants. The RPM curve on some motors deviates from the close to a
straight line suggested by the model. This is likely due to temperature effects as it is particularly noticeable on
the small motors that are close to being overloaded. The method of determining motor constants using
optimization to minimize the difference between the motor data and predicted data was less successful compared
to the previous test on the Mfly motor. Small changes in the test setup, such as two different propellers, yielded
too large variations in motor constants. The optimized constants would be too specific for that particular
operating condition, while lacking accuracy for other operating conditions. Generally, the optimized constants
always resulted in significantly higher Rm values in order to reduce efficiency to the measured values. However,
in order to still satisfy the RPM curve, the Kv values were increased. The I0 constant is the most difficult to
determine. On some motors, the optimized I0 values are unrealistically low. In some tests (not included in the
paper) the optimizer even found I0 to be zero or negative. This again seems to indicate that the original motor
model is not sufficiently accurate and that improvement, as suggested, is needed.
The discrepancy between measurements and predicted results raises the question of how motor constants are
determined. There are a few different methods to determine motor constants. The simplest method consists of
two steps. In the first step, I0 is directly measured when the motor is run at no load and fed with a fixed input
voltage. In the second step, the motor is run at two different voltages for which rpm and current are measured.
Using equation (2) and the measurements from these two operating points, Kv and Rm can be extracted. The
measurements in the second step can be made either with no load or when the motor is driving a propeller. In
order to see how the no-load characteristics of the motors correlate with the manufacturers’ motor constants,
each motor was tested without any propeller attached and then compared with calculations using equation (2).
The results are shown in Figure 29.
Validation of Models for… 15
12000 14000
20000
12000
10000
10000
8000 15000
8000
6000 10000
6000
4000
4000
Measured 5000
2000 Measured Measured
2000
Predicted Predicted Predicted
0 0 0
2 7 12 17 2 7 12 17 2 7 12 17
U(V) U(V) U(V)
5000 10000
Measured 5000 Measured Measured
5000
Predicted Predicted Predicted
0 0 0
2 7 12 17 2 4 6 8 10 12 14 2 7 12 17
U(V) U(V) U(V)
In these tests, the prediction and measurement coincide fairly well. It seems that motor manufacturers only use
no load measurement data to establish motor constants. This makes some sense, as it is the simplest form of
measurement. However, motor constants established this way have no guarantee to satisfy equation (1) or (3) in
the motor model.
0,9
0,8
0,7
0,6
0,5
0,4
Eta CC Phoenix10
0,3 Eta TGY Plush 6A
Eta TGY Plush 12A
0,2 Eta TGY Plush 40A
Eta YGE4A
0,1
Eta YGE12
0
0 10 20 30 40 50 60 70 80 90 100
Duty cycle length
Remarkably similar results except for the YGE12 and CC Phoenix 10 controllers that exhibit strange
discontinuous efficiency curves. All curves were taken under similar operating conditions with the same load
and voltage.
0,9 0,9
0,8 0,8
0,7 0,7
0,6 0,6
0,5 0,5
0,4 0,4
The PH10 exhibits very odd efficiency curves where the bowl-shaped discontinuity changes shape and position
depending on operating conditions. The YGE behaves more rationally.
0,9
0,8
0,7
0,6
Eta YGE12 Hacker 8V
0,5 Eta YGE12 A05 2510 14V
0,4 Eta YGE12 Himaxx p13 12V
Eta YGE z2205 p85 11V
0,3
Eta YGE12 Z1705 p8 10v
0,2
Eta YGE12 Mfly p9 8V
0,1 Eta YGE12 C2022 p6 12V
0
0 10 20 30 40 50 60 70 80 90 100
Duty cycle
The result is surprising. It is positive to see that the curve shape is relatively similar for all motors. This indicates
that it could be possible to establish a controller model where the curve shape is scaled based on the motor
parameters. Unfortunately, finding a correlation between the curve shapes and motor parameters has proven to
be difficult.
usage of such a model is limited. For design automation using optimization on COTS components, this means
that the user would have to establish their own motor constants by means of testing. This will complicate the
optimization process as establishing a large database of components is time consuming. Also, if testing needs to
be done anyway, optimization could just as well be carried out using a matrix of real test data, rather than a
model using identified constants.
The testing of different motors has shown that the motor constants (of a traditional model) provided by the
manufacturers does not provide reliable results. None of the motors showed measured efficiencies on the level
predicted using the original motor constants. Typically, the measured efficiencies were 5-10% lower than
predicted. The highest measured efficiency of any of the motors was 79% (Mfly motor). The best way to
determine motor constants for brushless motors appears to be from practical testing where all motor
performances are measured, including efficiency, and where curve fitting is done for the region of interest. The
motor constants will then represent coefficients in a polynomial curve rather than the electrical properties they
were originally defined to represent. Determining motor constants is best done on the entire operating envelop of
U and I. The second-best method is to optimize motor constants along RPM and power output curves for a fixed
voltage. The third tested method of optimizing motor constants based on test data from a test run with a single
propeller has resulted in slightly odd results, especially for I0, and is not a particularly good method of
determining motor constants. Either way, it is important that the constants are optimized to satisfy both the RPM
equation (2), and power output equation (1).
The testing of motor controllers has provided valuable data on efficiencies at part-throttle settings. It is
shown that both operating conditions and motor parameters influence the ESC’s efficiency. For most of the tests
the ESC’s efficiency shows a close to linear relationship for duty cycles of about 40-100%. In flight, throttle
values below 40% are unlikely to be used. For optimization purposes, it therefore seems that a simple linear
model for ESC efficiency could be used. The slope of the linear relationship, however, changes on different
motors and it has not been possible to find a relationship between this behavior and motor parameters. On some
motor controllers, strange discontinuous efficiency curves have been observed and there seems to be no obvious
reason why this happens. It is likely a software-related problem within the ESC. Modeling this behavior seems
very difficult but on the other hand is not very important. A better conclusion is that those controllers should not
be used for MAVs, or any other application where high efficiency is crucial. In order to finally derive a motor
controller model based on test results, a more systematic analysis must be conducted. However, the number of
tests that needs to be carried out in order to investigate how each motor parameter and operating condition
affects the ESC’s efficiency, quickly grows to unmanageable proportions. Considering that the software in the
ESC also impacts its efficiency, it may be impossible to identify a proper model. Investigating a method for
modeling an ESC based on test results will be the subject of future studies.
In the present work, the validation of propeller model has been omitted. In the design framework, the
propeller is modeled using traditional blade element analysis. Future work will also include comparisons
between this and other models with experimental propeller data. Propeller data will partly be acquired in-house
from wind tunnel testing but also from data published by other researchers9,10.
References
1 Lundström
D., Krus P., “Micro Aerial Vehicle Design Optimization Using Mixed Discrete and Continuous Variables”,
11th AIAA/ISSMO Multidisciplinary Analysis and Optimization Conference, Sept. 2006, Portsmouth, VA, USA.
2 Lundström D., Amadori K., Krus P., ”Distributed Framework for MAV Design Automation”, 46th AIAA Aerospace
Aerospace Sciences Meeting and Exhibit, Jan. 2009, Orlando, FL, USA.
4 Amadori, K., Lundström, D., Krus, P., “Evaluation of Automatically Designed Micro Air Vehicles and Flight Testing”,
48th AIAA Aerospace Sciences Meeting and Exhibit, Jan. 2010, Orlando, FL, USA.
5 Guzzella L., Sciarretta A., ”Vehicle Propulsion Systems”, Springer Verlag, Berlin Heidelberg, 2005.
6 MotoCalc website: www.motocalc.com
7 ElectriCalc website: www.slkelectronics.com
8 Medusa Research website: www.medusaresearch.com
9 Deters R. W., Selig S. M., ”Static Testing of Micro Propellers”, 26th AIAA Applied Aerodynamics Conference, Aug.
Applications”.,44th AIAA Aerospace Sciences Meeting and Exhibit, Jan. 2006, Reno, NV, USA