This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JSEN.2019.2931002, IEEE Sensors
Journal

Design and Implementation of a Compact Rotational Speed and Air

Flow Sensor for Unmanned Aerial Vehicles
Federico Fries1,4 , Shane Kyi Hla Win1 , Emmanuel Tang1 , Jun En Low4 , Luke Soe Thura Win1,4 ,
Pablo Valdivia y Alvarado1,2,3 and Shaohui Foong1,2,3,4

Abstract— Unmanned Aerial Vehicles (UAV) have gained

massive interest due to their wide range of applications. In
this work, we focus on the design of compact rotation speed
sensors (RSS) and air flow sensors (AFS) that can be integrated
into nature-inspired monocopters UAVs which generate lift by
spinning at high frequencies. Combining barometric sensors
and sensing elements embedded into soft elastomers, compact,
accurate, reliable, and low power sensors that can measure
rotation and air flow speed can be fabricated. Through numer-
ical analysis and design, it is possible to customize the sensor
to measure rotation speed or air flow at specific operating
ranges. The AFS was experimentally capable of measuring
airflow speeds up to 60m/s, a range in which anemometers
commonly saturate at. Consequently, the developed RSS can
measure frequencies up to 10Hz accurately, whereas inertia
measurement units (IMUs) saturate at 5.5Hz. Finally, the RSS
sensors were mounted on a monocopter prototype, to evaluate
and assess the sensors under actual flying conditions, where
the UAV’s movements were tracked by an onboard flight
controller’s (Pixracer) in-built IMU as well as a high accuracy
external optical motion capture system (OptiTrack). The RSS
proved to be a reliable alternative to IMUs for navigation
control of Monocopters.

I. INTRODUCTION
The interest in Unmanned Aerial Vehicles (UAV) has in-
creased exponentially due to their wide range of applications
[1]. A vast amount of work has been done in the area of
design and modeling. However, much of the research has
been focused towards control techniques and algorithms [2].
In this study, we concentrate on the design of sensory devices
Fig. 1. Design and description of rotation speed (RSS) and air flow sensors
for monocopter sensing and navigation. Monocopters are also (AFS). (a) Rotation speed sensor and (b) Air flow sensor. (i) Physical views
called free rotors, and that means the entire vehicle rotates and components. (ii) Force Diagrams. (iii) Side view of the sensors with
along its vertical axis, generating sufficient lift to take off and dimensions. (iv) Top view of the sensors with dimensions.
hover. As a result, monocopters reach high rotation speeds.
The popularity of monocopters has increased due to their low
cost and efficiency, given the simple mechanism consisting of [4]. Measuring rotation speed is a common problem in
two dynamic parts [3]. As with other UAVs, the monocopters the industry, mostly in power electronics for rotor posi-
can be designed according to the task to be fulfilled, and tion detection. Diverse methods have been implemented,
parameters such as weight, size, and rotation frequencies for instance, electrostatic [5] or magnetic sensors based on
differ between models. Hence the sensing requirement for giant magnetoresistance [6], sensorless techniques such as
each design would also differ. image processing algorithms [7], back electronic force and
The relevant dynamic states required for the controlled injection-based methods, nonetheless high currents during
flight of monocopters are rotating speed and heading angle startup stage of the motor affect measurements accuracy
[8]. Moreover, mechanical solutions include resolvers and
This project was supported by the SUTD-MIT International Design Cen-
tre (IDC), SUTD Digital Manufacturing and Design Centre (DManD) and encoders implemented by [9][10]. Alternatively, linear Hall-
Temasek Laboratories SUTD. 1 Engineering Product Development (EPD), Effect sensors combined with an extended Kalman filter
Singapore University of Technology and Design (SUTD). 2 SUTD Digital (EKF) represent a viable solution [8]. However, none of the
Manufacturing and Design Centre (DManD). 3 SUTD-MIT International
Design Centre (IDC). 4 Temasek Laboratories SUTD. Author by correspon- above techniques are suitable for monocopters due to their
dence [email protected] size, weight, price and signal processing requires powerful

1558-1748 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JSEN.2019.2931002, IEEE Sensors
Journal

microcontrollers. Inertial measurement units (IMU) and al-

titude and reference systems (AHRS) [11] are commonly
used for UAV control. Nonetheless, reliable gyros can be
very costly, such as the ADXRS649 by Analog Devices Inc.
(US$90 per unit) can measure rotations of 60Hz. However,
the resolution is reduced and can only measure one axis.
More affordable versions usually saturate at frequencies
below the 6Hz [4] which narrows the design spectrum of
the aircrafts.
An additional parameter that can enhance navigation con-
trol is the air flow. Pitot tubes are commonly used for
this purpose; they have proved to be highly sensitive to
air flow and effective detecting linear velocities. However,
monocopters (and quadcopters) generate irregular wind flows
while flying, preventing pitot tubes from accurate speed mea-
surement [12]. Additionally, due to their exposed design, they
can be clogged by dust, or water particles [13]. Alternative
methods for airflow sensing have been presented. A velocity
meter based on hall effect sensors for quadcopter navigation
was proposed by [7] and [14] was capable of measuring
ground effect through contact-resistance air flow sensors. Fig. 2. Operation principle of the RSS and AFS. (a)(i) RSS at initial
In previous work, whisker-like sensors have been investi- state measuring the atmospheric pressure P0 . (a)(ii) During rotation the
sphere is pulled away from the center counteracting P0 . (b)(i) AFS at initial
gated [15][16][17][18] to detect specific natural frequencies, state. (b)(ii) The elastomer is compressed due FD increasing the measured
that can be triggered by the flow of air or water. Nonetheless, pressure.
multiple sensors are needed to identify a wide range of
streams. None of the previously mentioned technologies are
designed for air flow measurement of rotating aircrafts such A. Sensor Design
as monocopters. For this project, we propose to adopt and A body in rotation is exposed to multiple forces in
advance the concept presented by [19] and implemented by different directions simultaneously (Fig. 1(ii)):
[20], which consist of the fabrication of soft tactile sensors  >
combining affordable MEMS barometric sensors and elastic F = FC FD Fg (1)
components. The interest in the manufacture of soft sensors with FC centrifugal, FD drag and Fg gravitational forces.
rose in the previous years [21]; however, such devices have However, it is possible to neglect specific forces through
mostly been integrated into soft robotics platforms overlook- meticulous design of the sensor to ensure the measurement
ing their potential in other fields. Some of the advantages of one targeted parameter such as rotation speed or air flow.
in elastic sensors are their durability, multi-functionality and Given the high rotation speeds and the low mass of the the
versatility [22], making them good candidates to be mounted silicone and sensing elements, the gravitational force can be
on monocopters which spin at high frequencies. ignored for both cases, therefore Fg = 0. At initial state the
The contribution of this paper is the design methodology barometric sensor will sense the force exercised through the
of lightweight and compact rotation speed and air flow atmospheric pressure P0 = 100kPa (Fig. 2(i)).
sensors (Fig. 1) that are capable of measuring a wide range 1) Rotation Speed Sensor (RSS): the high rotation speeds
of rotating speeds and flows accurately. Additionally, this and orientation of the sensor allow to ignoredrag FD and
methodology allows customization of the devices according >
gravitational Fg forces, as a result F = FC 0 0
to the specific parameters of each UAV. The implemented applies. The rotation frequency f is directly related to the
hardware, mathematical models, and fabrication procedures centrifugal force, hence
are presented in section II, and the experimental setups used
for this project are described in section III. The results of FC = α∗ Ms R(2πf )2 + β ∗ , (2)
the performed tests are presented and discussed in sections
where Ms is the mass of the sensing element, and R the
IV and V respectively and the study is concluded in section
distance from the center of rotation O to the sensor. Given
VI.
the characteristics of the polymer [24] and to compensate
fabrication imperfections a constant α∗ was included, while
II. SENSOR MODELING AND FABRICATION β ∗ represents an offset (both parameters calculated through
The sensors presented in this work are a combination experimentation). Furthermore, the force FC can be trans-
of affordable MEMS barometric sensors (MPL115A2, NXP lated to pressure as follows
Semiconductors) customized by Righthand Labs [23], and a α∗ Ms R(2πf )2 + β ∗
sensing element, embedded into a super soft elastomer. PC = , (3)
ARSS

1558-1748 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JSEN.2019.2931002, IEEE Sensors
Journal

with the base of the silicone ARSS = 42mm2 . The relevant

parameter to measure FC is the mass Ms , thus a stainless
steel sphere (4mm diameter and 256mg weight) was chosen
as sensing element. The high density of steel allows to embed
an element of similar area of the barometric sensor (Fig. 1
(a)(iv)) heavy enough to amplify FC . As a result, the total
mass Ms is 526mg (steel sphere and silicone together).
Additionally, the sensor is placed along the x-axis parallel
to FC maximizing sensitivity.
2) Air Flow Sensor (AFS): the AFS is placed perpendic-
ular to the air flow. Given the weight and thickness of the
sensing element the > force FC can be ignored,
 centripetal
therefore F = 0 FD 0 applies. analogous to the
rotation speed, the air flow is related to the frequency f
and can be measured through the drag force
1
FD = − γ ∗ ρair CD As (2πf R)2 , (4)
2
ρair being the density of air, CD drag coefficient (geometry
dependent), As area of the sensing element, γ ∗ scaling
parameter to overcome fabrication imperfections and, R the
distance between the center of rotation O and the AFS. As
a result the change in pressure due to the drag force can be
measured as follows: Fig. 3. The RSS and AFS fabrication process. (a) The barometric sensors
are fixed into a base. (b) Moulds with the shape of the Ecoflex 20 (EF20) are
∗ 2
γ ρair CD As (2πf R) mounted on the base (transparency was added to M11 and M21 for better
PD = − . (5) display) and first silicone layer is cured. (i) The sphere place holder is fixed
2AAF S to M11. (ii) M21 is placed on the base. (c) Sensing elements are covered
(i) The sphere replaces the placeholder. (ii) The cross is fixed between M21
For the AFS the sensing area As has to be maximized, while and M22.
mass must be kept low, therefore the sensing element would
be a 3D printed cross-like shape (Mark Two, Markforged).
The Mark Two is a 3D printer designed to print carbon 2) AFS: the drag force FD is amplified by the cross,
fiber parts. Carbon fiber is a lightweight and highly durable generating enough force to compress the silicone increasing
material, given the high resolution of the printer it was the pressure inside the elastomer (Fig. 2 (b)(ii)), thus a
possible to fabricate a sensing element with an area of negative signal is measured:
41.4mm2 , mass of 55.8mg and 300µm thickness, hence the
total sensing area As was 42.8mm2 . A cross was chosen to ∆PD = P0 − PD < 0 (7)
spread FD evenly avoiding torques. Carbon fiber was used
The deformations in within the elastomer can be neglected
for the part allowing to be lightweight, flexible and strong
(as shown in Figs. 8(c) and 12(c)), therefore it can be
while exposed at high air flow speeds. Any thinner part
assumed that the exercised forces are entirely transferred
would be too brittle for the sensor, especially at high speeds.
from the sensing element to the barometric sensor.
The contact area between the barometric sensor and the
elastomer has to exceed the barometric sensor dimensions, C. Fabrication
therefore AAF S = 28mm2 was set.
The task of the elastomer is to transfer the exercised
force to the MEMS barometric sensor. Hence the connecting
B. Measurement Principle layer between the sensor and the sensing element must be
The barometric sensor measures the relative pressure dif- minimized to avoid energy losses. The polymer (Ecoflex-
ference. At initial state the atmospheric pressure P0 (100kPa) 20, Smooth-On Inc.) was chosen due to its softness (Youngs
is recorded, which is the same inside the elastomer (diagrams Modulus E = 55kPa) and low viscosity (3000cps) [25]
in Fig. 2 (i)). The measured pressure will be the difference simplifying the fabrication process. The low viscosity of the
between the atmospheric and the exercised pressures. uncured material enables the fabrication of a thin membrane
1) RSS: during rotation the sensing element (steel sphere) (500µm) between the barometric sensor and the sensing
is pulled outwards due the centrifugal force FC generating element (Fig. 1(iii)).
pressure opposite to P0 (Fig. 2 (a)(ii)) resulting on a positive 1) RSS Fabrication: The sensors are fabricated in two
pressure difference: steps following shape deposition manufacturing (SDM)
methods (Fig. 3). Firstly, the barometric sensors are fixed
∆PC = P0 − PC > 0 (6) into a 3D printed base (Fig. 3(a)). A 3D printed mould

1558-1748 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JSEN.2019.2931002, IEEE Sensors
Journal

Fig. 4. Controlled speed experimental setup. The RSS is fixed to

a cantilever and rotated by a brushless motor at different frequencies
controlled through PWM and a microcontroller.

Fig. 5. Vertical wind tunnel. A linear wind flow is generated by a propeller

placed at the bottom of the tunnel. The speed is regulated through PWM.
(M11) is assembled on top of the base. Consequently, a The experiment is monitored by a motion capture system (OptiTrack). The
place holder with the shape of a sphere is fixed to the mould RSS sensor is mounted on the aircraft (Fig. 7(a)) at different distances R
from the center.
above the sensor at a distance of 500µm (Fig. 3(b)(i)). The
fabrication of thinner membranes (following this method)
is no longer reliable, resulting in membranes of different was used to control the different frequencies through pulse-
thicknesses compromising replicability of the samples. The width modulation (PWM). The sensor signal was recorded by
Ecoflex 20 (EF20) is mixed, degassed using a vacuum a microcontroller (Teensy 3.5, PJRC), mounted on the wheel
chamber, poured into M11, and cured in an oven at 60◦ C and controlled wirelessly by an infrared remote control. A
for 2 hours. Secondly, after the EF20 cured, the place holder magnet was glued to the bottom of the plate, allowing a Hall-
is removed and replaced by the sphere, and consequently effect sensor (HE) to detect each turn. Hence, the rotation
the second layer of EF20 is prepared following the previous frequency is set by the microcontroller. The sensor was
steps and poured into an additional mould (M12) to cover the linked to a rotating cantilever with a 3D printed mount.
sphere and let to cure in an oven (Fig. 3(c)(i)). The moulds The speed sensor was placed facing outwards parallel to
M11 and M12 are carefully separated from the base, and FC at a fixed radius of R = 135 mm from the center
the sensor is detached from the base. The resulting sensor is and tested at nine different frequencies ranging between 2Hz
presented in Fig. 1(a)(i), the total weight was 767.2mg. EF20 and 10Hz. The experiment was carried out three times, and
does not chemically bond to the PBC, hence, a thin layer of the data was recorded at 100Hz for a period of 15s. The
adhesive for silicone (Sil-Poxy, Smooth-On Inc.) was added experiment was recorded at a speed of 6400 frames/s by
between the PBC and the bottom of the elastomer to avoid a high-speed camera (FASTCAM Mini UX 100, Photron),
separation at higher speeds. placed in front of the experiment to see the deformation
2) AFS Fabrication: The manufacturing process is paral- of the silicone. An additional experiment was carried out
lel to the RSS sensor. The first step consists of fabricating to show the inhibition of drag forces through the design of
the connecting layer (also 500µm thickness) using a mould the sensors. The samples were covered with a 3D printed
(M21) (Fig. 3(b)(ii)). M21 contains cavities to place the box, preventing the air flow from touching the sensors. The
sensing element during the following step. The EF20 is experiment detailed above was repeated in this configuration.
mixed, poured into M21, and cured. Secondly, the cross- Finally, the test was performed with an inertial measurement
like shaped sensing element is placed on M21 and then fixed unit (IMU) available in the market (MPU9250, TDK) for
with an additional mould (M22). The second layer of EF20 is comparison. The gyroscope is not affected by external forces
prepared and poured into M22 covering the sensing element. (FC and FD ), therefore it is placed at the center of the plate
The resulting AFS is depicted in Fig. 1(b)(i) and has a total next to the microcontroller.
weight of 358mg.
B. Vertical Wind Tunnel
III. EXPERIMENTS AND PROCEDURES
A wind tunnel setup was built to test the sensor in an
A. Controlled Speed Measurement environment where the rotational speed is no longer directly
An experimental setup to test the RSS under controlled controlled (Fig. 5). The entire setup is approximately 2m
conditions was prepared (Fig. 4). A customized acrylic plate tall, with a structure made of aluminum extrusions. The tube
of a 160mm radius was designed and fixed to a brushless measuring 0.6m×0.8m (diameter×height) is made of clear
motor (U8, T-Motor) running at 24V. A servo tester (Turnigy) acrylic. A propeller (22×13in, Xoar) attached to a brushless

1558-1748 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JSEN.2019.2931002, IEEE Sensors
Journal

Fig. 6. Air flow experiment. Streams up to 60m/s of air are generated by

a dispenser. A hose is fixed in front of the AFS at 4.5cm. The experiment
was first calibrated by an anemometer.
Fig. 7. Aircrafts used for the experiments. (a) Top view of a two-wing
configuration aircraft designed for the vertical wind tunnel. The motion
capture system tracks the rotation through the movement of the markers.
motor (U8, T-Motor) at the bottom of the tunnel, running at (b) The monocopter (THOR) used for the real flight experiment. The
rotation was measured by the RSS, OptiTrack and flight controller (Pixracer)
24V provided sufficient air flow for the aircraft (Fig. 7(a)) to simultaneously.
spin. The rotation speed of the propeller was controlled with
PWM regulated by a servo tester (Turnigy). Plastic straws
fitted close after the propeller ensure uniform airflow inside was repeated for R = 85mm and R = 250mm.
the test region. A steel rod fixed in the middle of the tunnel
serves as a guide for the testing prototype to prevent from C. Direct Air Flow Measurement
hitting the side walls of the tunnel. The rod was lubricated to The setup to study the airflow sensors (AFS) is presented
avoid friction during the experiment. It is possible to achieve in Fig. 6 and consisted on a fluid dispensing control (Ultimus
up to 2.5 m/s vertical airspeeds. These vertically driven air V, Nordson EFD) able to generate air flows up to 60m/s by
flow will be used to induce rotation onto the spinning aircraft. controlling the pressure (between 10 to 99psi). Holders for
A two-wing configuration aircraft (Fig. 7(a)) was built to the sensor and hose (of 3mm diameter) were 3D printed
mount the sensors. The prototype consists of a 3D-printed and fixed to an optical table. The hose was placed in front
body, designed such that the hole in the center allows z- of the sensor at a distance of 4.5cm, allowing to start at
axis translation and rotation along the rod on the xy-plane low airspeeds. In the first step, the setup was calibrated,
perpendicular to the wind flow. Slots to fix a Teensy 3.5 and the generated flow was measured with a hot wire
for data acquisition and battery were included in the design. anemometer (RS Pro 1340, RS Components). Consequently,
The wings were made of balsa (2mm thickness) and laser cut. the anemometer was removed and the AFS was placed in
The aircraft was commanded to fly at differing frequencies at the same position. The pressure was varied by the dispenser
the same height ensuring the identical conditions throughout between 10 and 99psi in steps of 10 generating the air stream;
the experiment. The wind tunnel is placed inside an area, such pressures represent speeds between 10 and 60m/s. Each
monitored by a motion capture system (OptiTrack Prime speed was measured three times for a period of 10s. The
41, NaturalPoint Inc.). Hence, the rotational position of the sensor signals were recorded with a Teensy 3.5. The entire
aircraft could be tracked and recorded with high precision. setup was covered with a box to avoid air flow from the
The first experiment consisted of generating various wind environment. The experiment was recorded by the Photron
flows with the propeller. As a result, the aircraft rotates high-speed camera at 6400 frames/s.
at different speeds. The rotation speed was monitored with
the OptiTrack, while speeds between 2.5 and 6.5Hz were D. Application Example: Monocopter Flight
measured with the RSS at four different distances R (85mm, The performance of the RSS was tested on one of the
140mm, 195mm, and 250mm) from the rotating center O. monocopters built in our lab, THOR (Transformable Hov-
The response time of the RSS was investigated by generat- ering Rotorcraft). An updated version of the monocopter
ing transient wind profiles. Firstly the speeds were increased previously designed in our lab [26] was used for the exper-
from 2.5 up to 6.5Hz in steps of 1Hz resulting in a stair iment and is shown in Fig. 7(b). The prototype comprises
profile; the sequence was repeated reducing the speed. Sec- of two wings and motors (Snail, DJI) attached to it along
ondly, a step was created from 2.5 to 6.5Hz. The experiment with a flight controller (Pixracer, PX4 Autopilot) and a

1558-1748 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JSEN.2019.2931002, IEEE Sensors
Journal

power distribution system. The aircraft is 100cm wide and speed; each point represents the average mean of the mea-
weighs approximately 500g. The controller comes with an surements and their standard deviation. The mass Ms of
accelerometer, gyroscope, and compass to record navigation the sphere and silicone together was 526mg. The centrifugal
data (ICM-20608, TDK). The Pixracer also recorded the force is pulling the sphere away from the sensor, resulting
flight for comparison. The RSS was placed at 135mm from in positive pressure difference, dispersed along the entire
the center (same distance as in controlled speed experiment) silicone body. Therefore, the area ARSS equals the area of
the sensor is integrated with its power supply at the center the elastomer (42mm2 ). The initial pressure difference ∆PC
of the craft that is near the center of gravity to avoid was 0, confirming that Fg can be neglected. As predicted
imbalance to the aircraft upon taking off or cruising. The data by Eqs. (3) and (6) the difference grows exponentially by
acquisition system (Teensy 3.5) was mounted on the center increasing frequencies, furthermore, the measurements agree
and controlled through infrared remote control. The craft with the proposed model (solid line) for all frequencies.
spun at approximately 5.1Hz so that it could hover steadily The sensor was capable of accurately measuring the speeds
and not drift excessively during flying. The monocopters between 2 and 7Hz (highest deviation of 3.6%), which lies
flight path was controlled manually. Silver ball markers were in the range we are interested in. The deviation increases for
placed on the wings to enable the motion capture system frequencies above 7Hz. The images recorded by the high-
(OptiTrack) to track the crafts flight path, rotation speed speed camera show the deformation of the sensor during the
as well as altitude. The motion capture system traces the experiment. The sensor operating at 10Hz is presented in Fig.
markers using IR Cameras mounted near the ceiling. 8(b), while large deformation at a higher frequency (15Hz)
is visible in Fig. 8(c). In a second experiment, the sensor
IV. RESULTS was covered with a box to avoid drag forces. As shown in
A. Calibration Fig. 8(a) (green diamonds) the measurement was identical to
the previous test. It was possible to neglect the drag force
The constants α∗ , β ∗ , and γ ∗ were included to Eqs. (3)
through design ensuring the measurement of rotation speed.
and (5)) to compensate the imperfections caused during the
fabrication process and material properties. The constants α∗
and β ∗ were calculated with the data of the experiments, con-
trolled speed and wind tunnel, using least squares method.
The values for the RSS were:

α∗ = 1.1087 β ∗ = 0.0043kPa. (8)

Same method was applied to calculate γ ∗ with the direct air

flow experimental data:

γ ∗ = 0.9184 (9)

The proposed models were used to calculate the rotation

speed during the transient experiment and monocopter flight.
Fig. 9. Comparison of RSS (blue squares) and IMU (green circles). Both
sensors perform accurately until 5.5Hz where the IMU saturates. The RSS
is capable of measuring frequencies up to 10Hz accurately.

Finally, an IMU was tested under the same circumstances.

The comparison between the RSS and IMU is depicted in
Fig. 9. The x-axis describes the known rotation frequency
fknown measured by the setup’s microcontroller, whereas the
y-axis the measured frequency fmeas . The values measured
by the RSS are converted to fmeas using the model presented
Fig. 8. Controlled speed measurements. (a) The centrifugal force FC
in Eq. (3). Analogous to the RSS, the IMU performs highly
was measured through pressure P at different rotation frequencies (blue accurately until it saturates at approximately 5.5Hz, limiting
squares). In a second experiment, the RSS was covered to avoid drag forces the rotation frequencies of the monocopters. In contrast, the
FD (green diamonds). Each point represents the average mean of three
measurements and their standard deviation. (b) High speed camera image
RSS precisely records frequencies up to 10Hz.
at a rotation frequency of 10Hz. (c) High speed camera image at a rotation
frequency of 15Hz. C. Vertical Wind Tunnel
The results of the wind tunnel experiment are presented
in Fig. 10. The experiment was conducted three times for
B. Controlled Speed Measurement each speed; each point represents the average mean of the
The results of the first experiment are presented in Fig. measurements and their standard deviation. The rotation
8(a). The experiment was conducted three times for each speed of the aircraft was measured with the motion capture

1558-1748 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JSEN.2019.2931002, IEEE Sensors
Journal

system (Optitrack) for comparison. The sensor showed a D. Direct Air Flow Measurement
precise performance at all the positions. The measurements The dispenser used, allowed to test the sensor under
coincide with the predictions made by Eq. 3. It is possible extreme conditions. The maximum velocity was 60m/s
to observe that the air flow in the tunnel did not affect the (216km/h). The experiment was conducted three times; each
measurements and the accuracy of the sensors remained. data point represents the average mean of the measurements
The centrifugal force FC is proportional to the radius R, and their standard deviation. The drag area As = 58.8mm2
the further away from the center the more sensitive and less and the drag coefficient represents the one of a plate per-
stable. The average deviations of 85mm, 140mm and 195mm pendicular to the flow. The outcome of the experiment is
was 2.45%, whereas the deviation at the largest distance was displayed in Fig. 12. As predicted by Eqs. (5) and (6),
12%. the pressure difference decreases exponentially by increasing
speeds. The response resembles the behavior stipulated in Eq.
(5). The deviation increases at higher speeds considerably,
nonetheless the sensor is capable of measuring high speeds as
well. The compression of the AFS at 60m/s (99psi) is shown
in Fig. 12 (b)(c), and was measured with a video analysis tool
(Tracker [27]), the compression was approximately 0.5mm.
E. Application Example: Monocopter Flight
The RSS was mounted on a monocopter developed in our
lab (THOR). The craft was set to spin at 5.1Hz, and the flight
was manually controlled for 38s. The frequency was calcu-
lated with Eq. (3). The data was recorded by the OptiTrack,
Pixracer’s in-built IMU and the RSS simultaneously (Fig.
Fig. 10. Results of the RSS measurements mounted on a two-wing 13(a)). The aircraft accelerates and reaches the highest speed
configuration aircraft at different distances R in a vertical wind tunnel. after 5s and hovers at a constant altitude of approximately
The solid lines represent the mathematical model. Each point represents the
average mean of three measurements and their standard deviation. 5cm (Fig. 13(b)). The flown path is depicted in Fig. 13(c).
The average frequency during constant rotation (5s-35s)
measured by the RSS was 5.18Hz, while the OptiTrack and
The results of the transient experiments can be seen in Pixracer measured 5.15Hz and 5.13Hz respectively. Addi-
Fig. 11. Two different profiles were generated, stairs and tionally, the root-mean-square error (RMSE) of the Pixracer
step response. The the model proposed previously (Eq. (3)) and RSS relative to the Optitrack were calculated and were
and the calculated values in Eq. (6) were used to translate 0.028Hz and 0.085Hz respectively. The data of the three data
the measured pressure PC into rotation frequency f and sets were filtered in MATLAB using a median filter.
were compared with the OptiTrack signals. Both signals were
V. DISCUSSION
filtered in MATLAB using a median filter. The RSS did
not show lag during the acceleration and deceleration of the A. Fabrication
aircraft, and both measurements were taken at R = 80 mm The use of elastomeric materials for soft robotic applica-
and R = 250mm agree with the OptiTrack data. The RSS tions were discussed by [24]. Despite the mathematical mod-
also complies during constant flight. els available to describe the behavior of polymers, samples
might vary due to the material properties and the fabrication
processes. Therefore, the centripetal and drag force models
were adapted by including the constants α∗ , β ∗ and γ ∗ .
The values were calculated with experimental data using the
least mean square error. β ∗ was an offset while α∗ and γ ∗
were scaling factors varying by ±10% from 1. Therefore,
the models proposed proved to be highly accurate.
B. Rotation Speed Sensor (RSS)
1) Controlled Speed Experiment: our device was tested
in a controlled setup, without disturbances of the environ-
ment. The RSS was placed at 135mm from the center and
behaved as predicted by the proposed model (Eq. (3)). The
measurements were consistent and accurate between 0-7Hz
Fig. 11. Results of the transient experiments. Two different profiles were (maximal deviation of 3.6%). The deviations increased for
generated in the wind tunnel, stairs (top graphs) and step response (bottom).
The RSS was positioned at two different positions (85mm and 250mm)
higher frequencies, this can be attributed to the nature of the
elastic material and the additional layer of adhesive at the
bottom of the sensor. Visco-elastic materials do not behave

1558-1748 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JSEN.2019.2931002, IEEE Sensors
Journal

Fig. 12. The AFS was tested under extreme conditions using a dispenser.
(a) Measurement results. Each point represents the average mean of three
measurements and their standard deviation. (b) High-speed camera image
of the AFS before measurement. (c) Image of the sensor compression under
60m/s (99psi) air flows.

linearly when they are exposed to large stresses, which is the

case at higher rotation frequencies f (Fig. 8(c)).
2) Wind Tunnel: the fabricated prototype was mounted
on a two-wing configuration aircraft and tested in a wind
tunnel at differing rotation frequencies f (2.5-6.5Hz) and
radius R (85mm, 140mm, 195mm, and 250mm). The air
stream provokes the rotation of the device. Despite the Fig. 13. Frequency measurement during THOR flight. The signals were
recorded by OptiTrack, and RSS.(a) Rotation speed measured by the
external disturbances, the performance remained accurate different instruments. (b) Image of the monocopter flight. (c) Flight path
and reliable. The divergence at a larger R was more evident measured by Pixracer.
in this experiment, hence the deviations increased. The av-
erage deviations of 85mm, 140mm and 195mm was 2.45%,
whereas the deviation at the largest distance was 12%, due lower than 10m/s, occasioning irregular air flows between
to the nature of the elastic material. Mounting the sensor trials. Finally, the sensing element vibrates with increasing
at 250mm from the center also increased the moment of flows causing noise. The dispenser and anemometer saturate
inertia reducing the stability of the aircraft, as a result, the at such values, and it was not possible to experiment any
maximum frequency was 6Hz (instead of 6.5Hz). Sensors further. However, the AFS did not show signs of saturation.
for smaller monocopters, thus higher rotations speeds, can be Nevertheless, the flows to be measured during monocopter
fabricated by using a harder elastomer, for instance, Dragon flight lie below 20 m/s.
Skin (Smooth-On Inc.) with a 150kPa Young’s modulus.
The transient experiments with two different profiles (stair VI. CONCLUSION
and step) showed the response of the sensors during con- Rotation speed measurement is a common problem in
stant state and sudden speed changes. Additionally, the the industry, however current solutions such as gyroscopes,
least favorable scenarios were tested, 85mm showed less motion capture systems, anemometers, and pitot tubes do not
sensitivity, while 250mm reduced the stability of the aircraft. represent an optimal solution for UAV flight control. Thus,
Nonetheless, the frequencies were accurately measured using on this work, we focused on the design of affordable and
the proposed models. The discrepancy found at measure- compact rotation speed and air flow sensors for monocopters
ments 85mm from the center can be attributed to the lower (and UAVs) following the same principles. A mathematical
sensitivity of the sensor. model describing the behavior of the devices was established
allowing to customize the sensors according to the relevant
C. Air Flow Sensor (AFS) parameter (rotation or air flow) and the needs of the aircraft.
Airspeed is an important parameter, and it helps enhance Consequently, we proposed a simple and reliable fabrication
the navigation control of UAV and also contribute to a better methodology by combining affordable MEMS barometric
design of the wings. Following the proposed methodology sensors available in the market with sensing elements embed-
an AFS was fabricated by pushing manufacturing methods ded into a super soft elastomer such as EF20. Through the
(soft sensor fabrication and 3D printing) to the limit. The design of the device and implementation of sensing elements
sensor was capable of measuring speeds between 10-60m/s. such as a stainless steel sphere and a 3D printed cross-like
The model proposed in Eq. (7) described the behavior plate the centripetal FC and drag FD forces were measured.
of the sensor correctly. Nonetheless, the deviation of the A series of experiments were carried demonstrating the
measurements increases at higher speeds. Different factors correctness of the models and the performance of the sensors.
can cause such results. firstly, the material characteristics The controlled speed and wind tunnel experiments showed
already discussed above. Secondly, the hose was placed at accurate measurements of the frequency f up to 10Hz with a
4.5cm from the AFS to generate streams of air with speeds maximum deviation of 3.6% (deviations above 10% for R =

1558-1748 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JSEN.2019.2931002, IEEE Sensors
Journal

250mm). Additionally, the RSS recorded correctly despite [4] Teguh Santoso Lembono, Jun En Low, Luke Soe Thura Win, Shaohui
the airflow generated inside the wind tunnel. The results Foong, and U. Xuan Tan. Orientation filter and angular rates estima-
tion in monocopter using accelerometers and magnetometer with the
also provided valuable information regarding the placement Extended Kalman Filter. IEEE International Conference on Robotics
of the sensors. The highest accuracy and resolution was and Automation, pages 4537–4543, 2017.
found at distances 140mm and 190mm, ensuring accurate [5] Lijuan Wang, Yong Yan, Yonghui Hu, and Xiangchen Qian. Rotational
Speed Measurement Using Single and Dual Electrostatic Sensors.
measurements and stable flight of the UAVs. The response IEEE Sensors Journal, pages 1784–1793, 2015.
of the RSS was also demonstrated during the transient [6] C Giebeler, D J Adelerhof, A E T Kuiper, and J B A Van Zon. Robust
experiments at extreme scenarios, no lag and precise mea- GMR sensors for angle detection and rotation speed sensing. Sensors
and Actuators A: Physical, pages 16–20, 2001.
surements indicate their potential to be implemented for [7] Xiaochuan Zhang, Jinlong Chen, Zongtao Wang, Nan Zhan, and
UAV control. The RSS outperformed IMUs available in the Ricao Wang. Digital image correlation using ring template and
market, capable of measuring frequencies up to 10Hz, while quadrilateral element for large rotation measurement. Optics and
Lasers in Engineering, pages 922–928, 2012.
conventional sensors saturate at 5.5Hz. Finally, The sensors [8] Xinda Song, Jiancheng Fang, and Bangcheng Han. High-Precision
were mounted on the THOR, exposing them to real flying Rotor Position Detection for High-Speed Surface PMSM Drive Based.
conditions. The Pixracer, OptiTrack, and RSS measured the IEEE Transactions on Power Electronics, pages 4720–4731, 2016.
[9] Choong-hyuk Yim, In-joong Ha, and Myoung-sam Ko. A -to-
signals simultaneously, showing comparable results (5.11Hz, Digital Conversion Method for Fast Tracking. IEEE Transactions on
5.15Hz, and 5.18Hz respectively) throughout the entire flight. Industrial Electronics, pages 369–378, 1992.
Despite the movement and rotation of the aircraft, the sensors [10] Ali M Bazzi, Alejandro Dominguez-garcia, and Philip T Krein.
Markov Reliability Modeling for Induction Motor Drives Under Field-
recorded the signals accurately only 0.08Hz above the set Oriented Control. IEEE Transactions on Power Electronics, pages
speed (5.1Hz) and 0.03Hz above OptiTrack measurements. 534–546, 2012.
Furthermore, the RMSE of the RSS and IMU relative to [11] A Safaee, S Z Moussavi, M S Mehrabani, M B Menhaj, and
E Ghobadi. Construction and Control of Monocopter Using MEMS
the Optitrack were calculated (0.085Hz and 0.028Hz respec- AHRS. IEEE International Conference on Control and Automation
tively). In future work, the sensors will be used to control (ICCA), pages 219 – 224, 2014.
rotation speeds of monocopters above 5.5Hz. [12] Shady Zahran, Adel M Moussa, Sesay, Abu B, and Naser El-sheimy.
A New Velocity Meter Based on Hall Effect Sensors for UAV Indoor
Current sensors for air flow measurement do not represent Navigation. IEEE Sensors Journal, pages 3067–3076, 2019.
a viable option for monocopter flight. Hot wire anemometers [13] Søren Hansen, Mogens Blanke, and Jens Adrian. Diagnosis of UAV
are costly and bulky. Pitot tubes are accurate measuring linear pitot tube defects using statistical change detection. IFAC, 2010.
[14] Samuel Dutra Gollob, Yash Manian, Ryan St Pierre, Abraham Simp-
speeds, however, monocopters rotate to hover generating son Chen, and Sarah Bergbreiter. A lightweight , compliant , contact-
turbulent air flows, reducing the accuracy of pitot tubes resistance-based airflow sensor for quadcopter ground effect sensing.
significantly. Therefore, the flow sensor fabricated in this IEEE International Conference on Robotics and Automation, pages
7826–7831, 2018.
study represents a viable alternative to measure currents due [15] Federico Fries and Pablo Alvarado. Whisker-like sensors with soft
to the compact size and reliability. The AFS was subjected resistive follicles. IEEE International Conference on Robotics and
to different airspeeds and was capable of measuring speeds Biomimetics, pages 2038–2043, 2017.
[16] Pablo Valdivia Alvarado, Vignesh Subramaniam, and Michael Tri-
between 10 and 60m/s, behaving as predicted by the model antafyllou. Performance analysis and characterization of bio-inspired
(Eq. (7)). Additionally, the sensor did not show signs of whisker sensors for underwater applications. IEEE/RSJ International
saturation above 60 m/s, a range in which most anemometers Conference on Intelligent Robots ans Systems, pages 5956–5961, 2013.
[17] Pablo Valdivia y Alvarado and Sriharsha Bhat. Whisker-like sensors
saturate. In future work, we will use more sensitive MEMS with tunable follicle sinus complex for underwater applications. Pro-
barometric sensors to measure lower wind speeds. Moreover, ceedings - Bioinspiration, Biomimetics, and Bioreplication, 2014.
the sensors will be implemented for control. [18] Pablo Valdivia y Alvarado, Vignesh Subramaniam, and Michael Tri-
antafyllou. Design of a bio-inspired whisker sensor for underwater
To conclude, rotation speed and airflow are essential applications. Proceedings of IEEE Sensors, 2012.
parameters for navigation control of monocopters, the sen- [19] Yaroslav Tenzer, Leif P. Jentoft, and Robert D. Howe. The feel of
sors proposed outperformed current devices available and MEMS barometers: Inexpensive and easily customized tactile array
sensors. IEEE Robotics and Automation Magazine, pages 89–95, 2014.
represent an affordable, compact, accurate, low power and [20] Leif P. Jentoft, Yaroslav Tenzer, Daniel Vogt, Liu Jia, Robert J. Wood,
reliable solution to measure rotation and airspeed. As a result, and Robert D. Howe. Flexible, stretchable tactile arrays from MEMS
more sophisticated control algorithms could be implemented barometers. International Conference on Advanced Robotics (ICAR),
2013.
allowing the increase of rotation speed of the aircraft, ex- [21] Carmel Majidi. Soft Robotics: A PerspectiveCurrent Trends and
panding the range of monocopter capabilities and further Prospects for the Future. Soft Robotics, pages 5–11, 2014.
applications. [22] Daniela Rus and Michael T. Tolley. Design, fabrication and control
of soft robots. Nature, pages 467–475, 2015.
R EFERENCES [23] Righthand Robotics Inc. https://www.labs.righthandrobotics.com/.
[24] Jennifer C Case, Edward L White, and Rebecca K Kramer. Soft
[1] Bashi I. Dallal Omar, W. Z. Wan Hasan, N. Azis, S. Shafle, and Hiroaki Material Characterization for Robotic Applications. Soft Robotics,
Wagatsume. Unmanned aerial vehicle quadcopter: A review. Journal pages 80–87, 2015.
of Computational and Theoretical Nanoscience, pages 5663 – 5675, [25] Smooth - on. Ecoflex R Series, 2011.
2017. [26] Jun En Low, Luke Thura, Soe Win, Danial Sufiyan, Bin Shaiful,
[2] Sumaila Musa. Techniques for Quadcopter modeling and Design: A Chee How Tan, Gim Song Soh, and Shaohui Foong. Design and
Review. The Journal of Unmanned System Technology, 2017. Dynamic Analysis of a Transformable HOvering Rotorcraft. IEEE
[3] A Safaee, M S Mehrabani, M B Menhaj, V Mousavi, and S Z International Conference on Robotics and Automatation, pages 6389–
Moussavi. An experimental procedure for design and construction of 6396, 2017.
monocopter and its control using optical and gps-aided ahrs sensors. [27] D. Brown. Tracker video analysis and modeling tool.
International Journal of Mechanical and Mechatronics Engineering, https://physlets.org/tracker/.
pages 47–54, 2015.

1558-1748 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.