You are on page 1of 7

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/318698005

Modeling and control of a saucer type Coandä effect UAV

Conference Paper · May 2017


DOI: 10.1109/ICRA.2017.7989316

CITATIONS READS
3 77

5 authors, including:

Seung Hwan Song Hyouk Ryeol Choi


Sungkyunkwan University Sungkyunkwan University
4 PUBLICATIONS   10 CITATIONS    320 PUBLICATIONS   4,453 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Artificial muscle actuator View project

Research Fellowship (NRF- 2016R1A6A3A11933722) View project

All content following this page was uploaded by Seung Hwan Song on 02 November 2020.

The user has requested enhancement of the downloaded file.


2017 IEEE International Conference on Robotics and Automation (ICRA)
Singapore, May 29 - June 3, 2017

Modeling and Control of a Saucer Type Coandă Effect UAV


Jameson Y. Lee, Seung Hwan Song, Hyun Wook Shon, Hyouk Ryeol Choi, and Woosoon Yim

Abstract— The concept of utilizing the Coandă effect to complex, are widely modeled in the literature using blade
produce stable flight in saucer type Aerial Vehicles (AV) was element momentum theory (BEMT) [5], [6]. The effects of a
first conceived in 1935 by Henri Marie Coandă. Coandă’s mounted shroud have also been discussed in [7]–[9]. Shroud-
proposed AV design remained a relatively unexplored curiosity
for decades until the recent surge in popularity of small ing a propeller serves a dual purpose in that it both reduces
Unmanned Aerial Vehicles (UAV) revitalized interest in the the production of tip vortices and accelerates massflow via
topic. Coandă effect UAV offer some distinct advantages over a specific channel shape. The configuration of a Coandă
standard multirotors due to their robust frame design, which effect drone introduces an added complexity in that airflow
offers a unique mix of crash resistance, flight safety, and flight must be redirected through a large deflection angle over a
performance making them ideal for Human-Robot Interaction
(HRI) applications. While much work has been performed in broad convex hull. In a study related to this phenomena,
the area of characterization for Coandă surfaces and improved [10] discusses the propagation of turbulence in a fluid for
mechanical design of saucer type UAV, little work has been large angles of deflection in 90◦ pipe bends. The findings
found in the literature discussing stability and attitude control of this work imply a high correlation between changes to
of the platform. This work seeks to contribute to the study the shape of a fluid’s velocity profile during deflection and
of these drones by introducing an approximate servo mapping
for desired control effort of a particular prototype, and an pressure loss, which may impart a non-negligible effect on
experimental control law implemented within the framework lift concerning Coandă effect drones. Coandă surfaces do
of a modified commercially available flight stack. not generally increase lift due to skin effects, however, as
stated by [11] a Coandă surface may impart lift to a platform
I. I NTRODUCTION by redirecting air downwards via the convex shape of the
The first patent for a saucer type UAV design based on surface, or by entraining air from above the platform to create
the Coandă effect is credited to UK inventor Robert Collins a low pressure zone. The propeller/rotor pairing, shroud, and
in his 2003 patent no. 2387158 [1], [2]. In the following Coandă surface are highly coupled design elements for a
years several groups including Geoff’s Flying Saucer (GFS) Coandă effect drone, and determine the overall thrust and
project Ltd. and a Romanian academic consortium consist- torque production of the airframe at a given rotor speed.
ing of researchers from Galaţi Iaşi and Bacău universities The surfaces used for attitude control of a Coandă effect
developed prototypes with similar propulsion and stabiliza- UAV are treated in this paper as low aspect ratio flat plates
tion mechanisms [2]. GFS first introduced an octagonal in either a low or high angle of attack. The work of [12]
airframe with four roll/pitch flaps and radial yaw fins in specifically discusses forces induced on low aspect ratio flat
their 2007 work [3]. GFS project Ltd. would eventually plates at high angles of attack. It was found that the behavior
become Aesir Ltd., producing several more Coandă effect of a flat plate in free stream can be approximated using thin
UAV prototypes for military application including the ODIN, airfoil theory in low angles of attack, while at high angles
EMBLA, VIDAR, and HODER UAV described in [4]. In the plate undergoes a phenomena known as stall. In the stall
each design, at least one propeller or impeller is used to regime, pressure forces dominate aerodynamic effects, thus
induce airflow over a convex hull. The combined thrust normal forces induced on the plate tend to saturate.
produced by the propeller/rotor pairing and Coandă surface is In this paper, a dynamic model and controller for a
capable of generating lift to affect flight. The induced airflow prototype Coandă effect UAV in development at the Intel-
is then redirected over several control surfaces to affect thrust ligent Robotics and Mechatronics Systems (IRMS) Labo-
vectoring for attitude control. ratory at Sungkyunkwan University is presented. A Back-
A common element among the designs of Aesir is the stepping based control scheme is proposed within a case
use of a single shrouded rotor which induces airflow over study involving a dynamic simulation and experimental flight
a Coandă surface. Propeller dynamics, although generally tests. Of particular interest to this study is the development
of an accurate dynamic model for simulation and overall
*This work was supported in part by National Science Foundation’s PFI
Program(Grant No.1430328), 2016 NSF’s EAPSI program, NRF of Korea, flight performance in testing. Finally, a preliminary version
and the MSIP of Korea, under the G-ITRC support program (IITP-2016- of the control scheme was implemented on the prototype
R6812-16-0001) supervised by the IITP. Coandă effect UAV using a modified version of the ArduPilot
Jameson Lee and Woosoon Yim are with the Department of Mechanical
Engineering, University of Nevada Las Vegas, Las Vegas, NV, 89154, USA. firmware.
All correspondences should be delivered to Professor Woosoon Yim at
woosoon.yim@unlv.edu II. M ECHANICAL D ESIGN
Seung Hwan Song, Hyun Wook Shon, and Hyouk Ryeol Choi are with
the School of Mechanical Engineering, Sungkyunkwan University, Suwon, The saucer type S-Coandă drone in figure 1 was developed
16419, Republic of Korea. at the IRMS Laboratory. It utilizes a single rotor to induce

978-1-5090-4633-1/17/$31.00 ©2017 IEEE 2717

Authorized licensed use limited to: Sungkyunkwan University. Downloaded on November 02,2020 at 23:54:20 UTC from IEEE Xplore. Restrictions apply.
airflow over a Coandă surface. Because of the single rotor
design, the yaw fins shown in figure 1 are necessary to
compensate for the induced rotor torque on the propeller
due to drag. By changing the angle of attack of the yaw
fins using a servo, thrust vectoring about the body-fixed yaw
axis is achieved. The noted roll and pitch flaps function in
a similar manner, affecting thrust vectoring about the body-
fixed roll and pitch axes, respectively.

Fig. 2. Free body diagram of the IRMS S-Coanda drone. In this formulation
it is assumed that tangential drag effects on the control surfaces are much
less than the normal force effects. Ω denotes the angular velocity of the
propeller, and θi and γ denote angle of attack for the control surfaces relative
to some vector parallel with the body-fixed z-axis.

and z are body-fixed Cartesian coordinates, φ , θ , and ψ are


inertial frame angular displacements, p, q, and r are body-
fixed angular velocities, and Ω is the rotor angular velocity.
Lastly, Uδ defines the body-fixed control effort applied to
the system for δ ∈ {x, y, z, φ , θ , ψ}, and g is the gravitational
constant. To reach the result given in (1) it was necessary to
assume several conditions. In particular, the center of mass of
the platform lies on the origin of the body-fixed frame. The
x−axis defines front of the aircraft, and the y−axis defines
Fig. 1. Isometric view of the Intelligent Robotics and Mechatronics (IRMS)
the left of the aircraft. Gyroscopic, gravitational, and control
Laboratory S-Coandă drone prototype. The drone utilizes a single rotor to effort external effects are considered, however, disturbances
induce airflow over a Coandă surface, while using several control flaps to are neglected for simplicity. The angular velocity of the
stabilize its attitude.
propeller, Ω, is denoted positive in the CW direction about
the z+ −axis. (1) describes the dynamics of the Coandă effect
III. DYNAMIC M ODEL drone under the given assumptions.
To model the actuation modes of the IRMS saucer type
Ux Uy Uz
Coandă effect UAV, or S-Coanda, the basic free body di- Ẍ = cψ cθ + (cψ sθ sφ − sψ cφ ) + (cψ sθ cφ + sψ sφ )
agram of figure 2 was utilized. It is asserted that the pro- m m m
Ux Uy Uz
peller/rotor pairing, shroud, and Coandă surface collectively Ÿ = sψ cθ + (sψ sθ sφ + cψ cφ ) + (sψ sθ cφ − cψ sφ )
produce some aggregate force and torque about the plat- m m m
Ux Uy Uz
form’s body-fixed z−axis FP,C (Ω) and Manti (Ω), respectively, Z̈ = −g − sθ + cθ sφ + cθ cφ
m m m
where Ω is the rotor angular velocity. There are four flaps Iy − Iz Ip Uφ
used to maintain roll and pitch which induce some force upon ṗ = qr − qΩ +
Ix Ix Ix
the platform F(Ω, θi ) for some i ∈ {1, 2, 3, 4}, where θi is the
Iz − Ix Ip Uθ
ith flap angle. Each of the eight yaw fins move simultaneously q̇ = pr + pΩ +
Iy Iy Iy
and identically to affect yaw. The force applied to one fin
is assumed to be 18 Ff (Ω, γ) for some angle γ. As will be Ix − Iy Uψ
ṙ = pq +
discussed in the following sections of this paper, it was Iz Iz
assumed that the normal forces associated with each control (1)
surface are much larger than the tangential forces. The control effort mapping in (2) requires a specific set
of assumptions concerning both flow condition and model
A. Newton-Euler Formulism choice. Firstly, the flap angles θ1 , θ2 , θ3 , and θ4 correspond
The dynamic equations were derived using the Newton- to the control flaps located at the front, right, back, and left
Euler formulism, where c(·) and s(·) represent cos(·) and of the aircraft, while the fin angle γ corresponds to all yaw
sin(·), for some angle (·). m was used to denote the mass fins located radially about the aircraft. It is assumed that the
of the UAV, I p is the bulk inertia of the rotor about the center of pressure for each control surface is positioned at a
positive body-fixed z−axis, and Ix , Iy , and Iz are diagonal radius r from the body-fixed z−axis. It is also necessary
elements of the UAV inertia tensor, where the inertia tensor to assume that the variation in location of the center of
is assumed diagonal, that is Ixy = Iyx = Iyz = Izy = Ixz = Izx = 0. pressure for each of the four roll/pitch flaps is negligible,
The states of the system were defined by convention, where and that some plane P defined by the intersection of these
X, Y , and Z are inertial frame Cartesian coordinates, x, y, points is coincident to the body-fixed xy−plane. To further

2718

Authorized licensed use limited to: Sungkyunkwan University. Downloaded on November 02,2020 at 23:54:20 UTC from IEEE Xplore. Restrictions apply.
simplify the model, the tangential components of force
FT (Ω, θi ) and FT, f (Ω, γ) acting upon each control surface is 
αCNs +CN0 0 ≤ |α| < αNstall
assumed to be much less than each surface’s corresponding CN (α) =
normal components FN (Ω, θi ) and FN, f (Ω, γ). The general sgn(α)CNstall αNstall ≤ |α| < π2
 (4)
control surface forces acting upon the roll/pitch flaps and γCN, fs +CN, f0 0 ≤ |γ| < γN, fstall
yaw fins F(Ω, θi ) = FN (Ω, θi ) + FT (Ω, θi ) and Ff (Ω, γ) = CN, f (γ) =
sgn(γ)CN, fstall γN, fstall ≤ |γ| < π2
FN, f (Ω, θi ) + FT, f (Ω, γ), respectively, can then be simplified
to F(Ω, θi ) = FN (Ω, θi ) and Ff (Ω, γ) = FN, f (Ω, θi ). A control B. Derivation of the Control Law
effort mapping satisfying these stated assumptions is given
in (2). Like many VTOL-UAV Coandă effect drones possess
coupled position and attitude dynamics. Because of this, the
input-affine subsystem of (5) was considered in place of (1).
These state dynamics represent the altitude and attitude of
Ux = FN (Ω, θ3 ) cos θ3 − FN (Ω, θ1 ) cos θ1
the drone, where g : Rn → Rn×n and the drift term f : Rn → Rn
Uy = FN (Ω, θ2 ) cos θ2 − FN (Ω, θ4 ) cos θ4 are derived from (1).
n
Uz = FP,C (Ω) − FN, f (Ω, γ) sin γ − ∑ F(N Ω, θi ) sin θi
i=1 (2) η = f (η
η̇ η ) + g(η
η )U
U
Uφ = rFN (Ω, θ2 ) sin θ2 − rFN (Ω, θ4 ) sin θ4  T
U = UZ Uφ Uθ Uψ (5)
Uθ = rFN (Ω, θ1 ) sin θ1 − rFN (Ω, θ3 ) sin θ3  T
η = Ż p q r
Uψ = rFN, f (Ω, γ) cos γ − Manti (Ω)
Using the deterministic control law presented in (6),
The definitions for roll/pitch flap normal force FN (Ω, θi ), asymptotic convergence of the state vector η to some desired
yaw fin normal force FN, f (Ω, γ), bulk rotor force FP,C (Ω), trajectory η ∗ can be proven using Lyapunov principles,
where Λ(η η ) represents desired dynamics in terms of state
and bulk rotor torque Manti (Ω) are given in (3), assuming that
η ∗ (t), and positive
η (t), derivative of desired trajectory η̇
error η̃
airflow throughout the airframe is incompressible. It should
be noted that the propeller is modeled using BEMT, and that definite constant matrices α 1 and α 2 . It is assumed that all
all control surfaces are modeled using modern airfoil theory. states are measurable.

 
U ∗ = g−1 (ηη ) − f (ηη ) + Λ(η η)
1
FN (Ω, θi ) = CN (θi )ρaN Ω2 AN = c∗1CN (θi )Ω2 Λ(η η ∗ (t) − (α
η ) = η̇ α 1 + α 2 )η̃
η (t)
2  t
1 α 2 α 1 + I4×4 )
− (α η (τ)dτ + η̃
η̃ η (0) , α 1 , α 2 > 0
FN, f (Ω, γ) = CN, f (γ)ρaN, f Ω2 AN, f = c∗2CN, f (γ)Ω2 (3) 0
2
(6)
FP,C (Ω) = c∗0 Ω2
Manti (Ω) = c∗3 Ω2 By considering the trial Lyapunov function V (η̃ η , η̃
η˙ ) =
1 ηT η T
2 η̃ η̃ + 2 (−η̃
1 η − α 1 η̃
˙ η ) (−η̃η − α 1 η̃
˙ η ) > 0, expansion of
The incompressible flow assumption can be made for the time derivative Lyapunov function results V̇ (η̃ η , η̃
η˙ ) =
flow with Ma < 0.3, and while the airflow over the control −η̃η α 1 η̃
T
η − (−η̃ ˙ η ) α 2 (−η̃
η − α 1 η̃ T ˙ η ) < 0. This implies
η − α 1 η̃
surfaces is not identically uniform free stream, it was posited asymptotic convergence of state onto the desired trajectory,
at the outset of this study that approximate attack angle η → 0 as t → ∞.
or η̃
dependent normal lift coefficients CN (θi ) and CN, f (γ) could The control effort mapping is a highly nonlinear rela-
be used to model the behavior of the system. In (3) c∗0 , c∗1 , c∗2 , tion. For expedience in calculation, it was assumed that
and c∗3 are constant coefficients which would be determined the time between samples Δt is sufficiently small, such
experimentally in this study, where aN and aN, f are constant that the changes in position for the control surfaces are
coefficients used to correlate square angular velocity of the sufficiently small from sample to sample. In the case of
rotor Ω2 to square free stream velocity over each control the rotor effort control law (7), the terms representing
surface, ρ is the density of air, and AN and AN, f are span control fin and flap drag opposing thrust effort are as-
areas for both the roll/pitch control flaps and yaw fins, sumed to change relatively slow due to their mechanical re-
respectively. sponse. That is, c∗2CN, f (γ k ) sin γ k ≈ c∗2CN, f (γ k+1 ) sin γ k+1 and
c∗1 ∑ni=1 CN (θik+1 ) sin θik+1 ≈ c∗1 ∑ni=1 CN (θik+1 ) sin θik+1 imply
Based on the work of [12], the control surface normal
the validity of (7) for calculation of the k + 1 rotor effort.
lift coefficients were modeled as the piecewise continuous
definitions given in (4), where CNs , CN0 , CNstall , CN, fs , CN, f0 ,
and CN, fstall are constant, and the attack angles αNstall and Uz
(Ω2 )k+1 ≈
γN, fstall correspond to the minimum angle of attack for the c0 − c2CN, f (γ ) sin γ − c∗1 ∑ni=1 CN (θik ) sin θik
∗ ∗ k k
control surfaces to undergo stall phenomena. (7)

2719

Authorized licensed use limited to: Sungkyunkwan University. Downloaded on November 02,2020 at 23:54:20 UTC from IEEE Xplore. Restrictions apply.
To calculate the desired attack angles for the roll/pitch
control flaps two parameters ξ and ν are introduced to avoid
an overactuation condition, where θ1 = π4 + ξ , θ2 = π4 − ν,
θ3 = π4 − ξ , and θ4 = π4 + ν. This transform reduces the
roll and pitch effort to one degree of freedom each, and
forces neutral operation of the control flaps well within
the stall regime, thus validating the assertion that control
surface normal force is much larger than control surface
tangential force. The approximate motor mapping for ν k+1
and ξ k+1 in (8) and (9), are calculated based on the normal
lift coefficients CN (θik ) and previous rotor output (Ω2 )k .
√  
− 2
ν k+1 CN (θ2k ) +CN (θ4k )
tan ≈ √  2
 Uφ
2
2 CN (θ2 ) −CN (θ4 ) + rc1 (Ω2 )k
2 k k

2 Fig. 3. The body of the S-Coanda was mounted to the depicted test rig.
U
(8)
CN2 (θ2k ) +CN2 (θ4k ) − rc (Ωφ 2 )k An 8008 Combo tachometer from AZ Instruments was used to determine
1 rotor RPM for a given measured thrust and torque.
± √   Uφ
2 CN (θ2 ) −CN (θ4 ) + rc∗ (Ω2 )k
2 k k
1

√  
ξ k+1 2
CN (θ1k ) +CN (θ3k )
tan ≈ √  2
 Uθ
2
2 CN (θ1 ) −CN (θ3 ) + rc (Ω2 )k
2 k k


1
2 (9)
CN2 (θ1k ) +CN2 (θ3k ) − rc U(Ωθ 2 )k
1
± √   Uθ
2 CN (θ1 ) −CN (θ3 ) + rc∗ (Ω2 )k
2 k k
1

The attack angle control parameter for the yaw fins γ k+1
was calculated in (10) by reconciling acceleration effects
from both desired thrust and yaw control equations.
Fig. 4. An ATI nano force/torque sensor was mounted to the base of the
S-Coanda. The thrust and torque were measured for a given rotor effort,
Uz − c∗0 (Ω2 )k + c∗1 (Ω2 )k ∑ni=1 CN (θik ) sin θik where the body of the drone is depicted in figure 3.
tan γ k+1 ≈ −r
Uψ + c∗3 (Ω2 )k
(10)
An ATI nano F/T sensor was mounted to the base of the
IV. C ASE S TUDY drone, and the thrusting force and yaw torque were measured
The case study discussed in this paper involves the exper- as a function of rotor angular velocity. The results of this test
imental validation of the aerodynamic parameters presented were used to create the plot of figure 6. With the measured
in the model derivation. Using the validated parameters, data the model parameters could be calculated as in (11).
a dynamic simulation of the system was created, and a
preliminary version of the developed motor mapping was
FP,C (Ω)
implemented on a real-time flight controller for testing. c∗0 = ≈ 9.1189 × 10−6 (kg · m)
Ω2
A. Experimental Validation of Model Parameters 1
c∗1 = ρaN AN ≈ 2.5336 × 10−6 (kg · m)
The method and results for determination of the model 2 (11)
parameters c∗0 , c∗1 , c∗2 , and c∗3 are discussed in this section. 1
c2 = ρaN, f AN, f ≈ 5.5608 × 10−6 (kg · m)

The experimental setup for measuring flow velocity, thrust, 2
∗ Manti (Ω)
and torque as a function of rotor speed is given in figures 3 c3 = ≈ 1.3678 × 10−7 (kg · m2 )
and 4. In figure 3, a 8008 Combo tachometer manufactured Ω2
by AZ Instruments was clamped above the rotor assembly Lastly, the work of [12] was used to identify the normal
to measure each stroke of the propeller blade in RPM. lift coefficients necessary for simulation. The experiments
A TES-1340 hot-wire anemometer was used to locally performed in this work are extensive in type and parame-
approximate the average flow velocity directly above the terization, and possess lift coefficient data for flat plates of
roll/pitch flaps and yaw fins. The flow speed plot given similar aspect ratio (AR) to those of the Coandă effect drone
in figure 5 was generated using discrete measurements of control surfaces with similar Reynolds flow (Re ≈ 6 × 104 ).
steady-state flow for various rotor angular velocities. A linear The authors parameterized lift coefficients at both low and
relation was used to fit the data with high correlation. high angles of attack in a survey of many methods.

2720

Authorized licensed use limited to: Sungkyunkwan University. Downloaded on November 02,2020 at 23:54:20 UTC from IEEE Xplore. Restrictions apply.
Position of the S−Coanda
5

Inertial Frame Pos. (m)


4

1
X Pos.
0 Y Pos.
Z Pos.
−1
0 5 10 15
Time (s)
Attitude of the S−Coanda

Body−Fixed Frame Pos. (rad)


0.3

0.2

0.1

0
φ Pos.
−0.1 θ Pos.
ψ Pos.
−0.2
Fig. 5. Linear curve fit of measured airflow speed directly above both the 0 5 10 15
roll/pitch flaps and yaw fins as a function of rotor angular velocity Ω. The Time (s)
r2 value for the control flap airflow velocity curve fit is r2 = 0.9599, and
the r2 value of the yaw fin airflow velocity curve fit is r2 = 0.9731. Fig. 7. The position and attitude outputs from the case study simulation are
presented. A pseudo-real-time PID loop was used to generate φ θ reference
trajectories for a given desired XY Zψ trajectory. The Backstepping scheme
ensured convergence of altitude, roll, pitch, and yaw states to the desired
trajectory.

cussed as a means of stabilizing the system, any stabilizing


controller could have been used, since the mechanics of
control for any rotor type UAV based on accelerometer,
magnetometer, and gyroscope data will function in a largely
similar manner. The critical focus of analysis for this control
scheme was directed towards the motor mappings for control
surface angle of attack. The mapping used is approximate,
and relies on a high frequency update law to maintain
accuracy. The output for the simulation case study suggested
a maneuver of the type described by figure 7 is well within
the span of controllability for the system with no apparent
Fig. 6. Quadratic curve fitting of measured force and torque generated by saturation effects.
the platform using an ATI nano F/T sensor as a function of rotor angular
velocity Ω. The r2 value of the force and torque fit are r2 = 0.9884 and C. Implementation of the Controller
r2 = 0.9899, respectively.
The developed controller was implemented on the IRMS
experimental platform using the Pixhawk flight controller
and a customized APM flight stack. The Pixhawk flight
B. Dynamic Simulation
controller was chosen for this project due to its highly
A dynamic simulation of the model and controller was developed interface, and its ability to outperform nearly all
constructed in the MATLAB toolbox Simulink using the de- other commercial flight controllers at the time of this writing.
rived model (1), control law (6), and control effort mappings The Pixhawk utilizes a Cortex M4F CPU at 168MHz, and
(7), (8), (9), and (10). possess 256kB of RAM and 2MB Flash storage. This is a
In the presented simulated state data of figure 7, the marked improvement over previous controllers including the
Coandă effect drone was given 4th −order continuous piece- APM board which operates at 16MHz.
wise trajectories of position, velocity, and acceleration for Concerning the firmware itself, the ArduPilot project is
the X, Y , Z, and ψ states. A PID controller was used to written in C++ and supports several overarching frame types
generate the required roll and pitch reference trajectories as including fixed wing plane, rover, antenna tracker, and many
input into the discussed Backstepping controller in (6) based copter type UAV. The Coandă effect drone falls under the
on position error. This two part feedback scheme reflects copter frame category, however no native frame type for
the physical requirements of VTOL-UAV, which require the Coandă effect drone exists within the code base. The
specific external sensors for position control removed from ArduPilot project is versatile in that the vehicle specific
the generic attitude controller present in most commercial directories allow for any number of flight modes, mission
flight controllers. directives, and frame types to be added without drastic
While the Backstepping controller given in (6) was dis- code alteration. For this project it was necessary to add a

2721

Authorized licensed use limited to: Sungkyunkwan University. Downloaded on November 02,2020 at 23:54:20 UTC from IEEE Xplore. Restrictions apply.
new frame type to the common libraries directory shared mapping used to generate the desired control effort was
by all vehicle specific directories. For initial testing of the proposed. Using data collected experimentally, design coef-
developed motor mappings in (7), (8), (9), and (10), the ficients corresponding to the developed IRMS platform were
developed PID based APM attitude controller files were calculated for use in a dynamic simulation. Lastly, a prelim-
utilized to output a desired roll, pitch, yaw, and throttle, inary version of the motor mapping was implemented on a
while our preliminary motor mapping was implemented in Pixhawk flight controller using a custom APM flight stack.
new Coanda frame descriptor files in the motors library As this work progresses, the implementation of the motor
[13]. In this way, the desired thrust and torque for control mapping will be improved for increased stability, and the
can be affected using the proper servo commands. Several proposed Backstepping controller will be tested. Successful
successful flight tests were performed using a preliminary development of the custom flight stack may also result a
version of the motor mapping, and a still of the prototype format conforming derivative of the ArduPilot Project allow-
mid-flight was captured in figure 8. ing other developers to experiment with the Coandă effect
drone firmware, increasing the pool of potential developers
and researchers to forward the research.
ACKNOWLEDGMENT
This work was supported in part by National Science
Foundation’s PFI Program(Grant No.1430328), 2016 NSF’s
EAPSI program, NRF of Korea, and the MSIP of Korea,
under the G-ITRC support program (IITP-2016-R6812-16-
0001) supervised by the IITP.
R EFERENCES
[1] I. Circiu and S. Dinea, “Review of applications on coand effect.
history, theories, new trends,” 1935.
[2] O. Crivoi, I. Doroftei, and F. Adascalitei, “a Survey on Unmanned
Aerial Vehicles Based on Coanda Effect,” pp. 338–344.
Fig. 8. The prototype Coandă effect drone captured mid-flight using the [3] F. NEDELCU, “Coanda Effect UAV A New Born Baby In The
augmented APM flight stack. Unmanned Ariel Vehicles Family,” The Scientific Informative
Review, vol. 2, no. 17, p. 21, 2010. [Online]. Available:
Due to lack of position-feedback, the prototype Coandă http://www.afahc.ro/revista/Nr 2 2010/nr 2 2010.pdf#page=22
[4] F. N. , “Towards a New Class of Aerial Vehicles Using the Coanda
drone was piloted manually for initial testing. The attitude Effect.”
data given in figure 9 was logged over a 10 second flight in [5] T. Bresciani, “Modelling , Identification and Control of a Quadrotor
which the drone successfully launched, hovered, and landed. Helicopter,” Ph.D. dissertation, Lund University, 2008.
[6] S. Bouabdallah, “Design and Control of Quadrotors
While slight oscillations were observed during take-off and With Application To Autonomous Flying,” Techniques, vol.
landing, roll and pitch were maintained throughout the test. 3727, no. 3727, p. 61, 2007. [Online]. Available:
http://biblion.epfl.ch/EPFL/theses/2007/3727/EPFL TH3727.pdf
Attitude Log of Flight Test [7] V. Hrishikeshavan, J. Black, and I. Chopra, “Design and
0.5
Testing of a Quad Shrouded Rotor Micro Air Vehicle in
Hover,” 53rd AIAA/ASME/ASCE/AHS/ASC Structures, Structural
0 Dynamics and Materials Conference, 2014. [Online]. Available:
http://dx.doi.org/10.2514/6.2012-1720
[8] R. van Manen and M. Oosterveld, “Analysis of Ducted-Propeller
-0.5
Design,” in The Society of Naval Architects and Marine Engineers,
Attitude (rad)

New York, 1966, pp. 522–562.


-1 [9] J. L.Pereira, “Hover and Wind-Tunnel Testing of Shrouded Rotors for
Improved Micro Air Vehicle Design,” p. 349, 2008.
[10] K. Beij, “Pressure losses for fluid flow in 90 degree pipe bends,”
-1.5 Journal of Research of the National Bureau of Standards, vol. 21,
no. 1, p. 1, 1938.
-2
[11] C. Barlow, D. Lewis, S. D. Prior, S. Odedra, M. A. Erbil,
Roll
Pitch M. Karamanoglu, and R. Collins, “Investigating the use of the
Yaw coanda effect to create novel unmanned aerial vehicles.” Design, pp.
-2.5 386–391, 2009. [Online]. Available: http://eprints.mdx.ac.uk/3870/
0 2 4 6 8 10 12
[12] X. Ortiz, D. Rival, and D. Wood, “Forces and moments on flat plates
Time (s) of small aspect ratio with Application to PV wind loads and small
wind turbine blades,” Energies, vol. 8, no. 4, pp. 2438–2453, 2015.
Fig. 9. The Pixhawk flight controller log outputs for calculated body-fixed [13] “ArduPilot/ardupilot.” [Online]. Available:
roll, pitch, and yaw. The maximum angular displacement of roll and pitch https://github.com/ArduPilot/ardupilot
was maintained between 7◦ and −7◦ , while a controlled yaw was recorded [14] H. Lee, S. Han, H. Lee, J. Jeon, C. Lee, Y. B. Kim, S. H. Song,
in this hover test. and H. R. Choi, “Design Optimization, Modeling and Control of
Unmanned Aerial Vehicle Lifted By Coand Effect,” IEEE/ASME
Transactions on Mechatronics, vol. 4435, no. c, pp. 1–1, 2017.
V. C ONCLUSIONS [Online]. Available: http://ieeexplore.ieee.org/document/7833100/
In this paper a dynamic model for a particular Coandă
effect platform was presented, and an actuation or motor

2722

Authorized licensed use limited to: Sungkyunkwan University. Downloaded on November 02,2020 at 23:54:20 UTC from IEEE Xplore. Restrictions apply.
View publication stats

You might also like