Theoretical and Experimental Investigation

into the flight of an X-Zylo

Nils Wagner∗
arXiv:2102.02647v1 [physics.class-ph] 26 Jan 2021

Abstract
Flying Gyroscopes are fascinating flight objects, which, due to gyroscopic
stabilization, can achieve surprisingly long flight distances when thrown with
rapid spin. The most common example hereby is a traditional Frisbee disc.
This paper focuses on a similar object called X-Zylo, that shows a remarkable
straight flight despite its simple geometry.
The main aim of the present study is to investigate the flight behavior of the
X-Zylo and to build a reliable groundwork for further quantitative parameter
studies on ring wing configurations. To achieve this goal, a six degree of
freedom model to predict the flight trajectory was developed. The trajectory
computation uses interpolated high-fidelity CFD simulation data to calculate
the acting moments and forces on the object during flight. A launch contraption
was built to be able to validate the theory systematically and reproducible in
experiments without human factors involved in the launch.
Despite the complexity of the flight, the theoretical simulations match the
real world data qualitatively, however quantitative differences still prevail. The
investigation shows that the deviation between theory and experiment mostly
stems from uncertainties in the CFD data as well as the optical recording of the
experimental data. Despite the methods outperforming those of prior studies,
advancements still have to be made in those areas in order to obtain better
quantitative accordance between theory and experiment.

Keywords: trajectory simulation, CFD, ring wing, annular airfoil, toy aerodynamics

∗
This work begun as a project for the national science fair in Germany (Jugend forscht) and was later overhauled
for publication. Email address for correspondence: [email protected]
Contents
1 Introduction 1

2 Theory of Flight 2
2.1 First Drop and Equilibrium Phase . . . . . . . . . . . . . . . . . . . . . . . 2
2.2 Late Flight and Second Drop . . . . . . . . . . . . . . . . . . . . . . . . . 3

3 Trajectory Calculation 4
3.1 Basic Mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
3.2 Aerodynamic Forces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
3.2.1 Theoretical Approximations . . . . . . . . . . . . . . . . . . . . . . 6
3.3 Computational Fluid Dynamics . . . . . . . . . . . . . . . . . . . . . . . . 7
3.3.1 Geometry and Mesh . . . . . . . . . . . . . . . . . . . . . . . . . . 7
3.3.2 Simulation Settings . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
3.3.3 Validation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
3.3.4 Mesh Independence Study . . . . . . . . . . . . . . . . . . . . . . . 11
3.3.5 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
3.3.6 Influence of the Model’s Rotation . . . . . . . . . . . . . . . . . . . 13

4 Launch Construction 14
4.1 Explanation of the used Mechanism . . . . . . . . . . . . . . . . . . . . . . 15
4.2 Reproducibility of the Device . . . . . . . . . . . . . . . . . . . . . . . . . 15

5 Experimental Procedure and Trajectory Evaluation 17

5.1 Preparatory Measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
5.2 Camera Setup and Corrections . . . . . . . . . . . . . . . . . . . . . . . . . 17

6 Comparison between Theory and Experiment 19

6.1 Detailed Results for a single Launch . . . . . . . . . . . . . . . . . . . . . . 19
6.2 Influence of the Launch Angle . . . . . . . . . . . . . . . . . . . . . . . . . 26
6.3 Sideways Drift . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
6.4 Human Induced Launch . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
6.5 Open Issues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

7 Conclusion 33

8 References 34

9 Appendix 35
9.1 Error Approximation for the Simulation . . . . . . . . . . . . . . . . . . . 35
9.2 Error Approximation for the experimental Trajectory Observation . . . . . 36
Theoretical and Experimental Investigation into the flight of an X-Zylo Nils Wagner

1 Introduction
Besides paper airplanes several flying toys have been developed, which, thrown the right way, can
travel long distances through the air in a controlled manner. The “X-zyLo™” (from now on simply
called X-Zylo) is a lesser-known example of such toys. Thrown like a football spinning along its
long axis can result in flight distances up to 100 m and above. The distance record given by the
manufacturer is 655 ft or 199.6 m [1].

Relevant Data:

mass: m = (22.73 ± 0.16) g

length: l = 48.0 − 61.0 mm (lavg = 54.5 mm)
diameter: douter = 97.0 mm
thickness: 1 mm (leading edge), 0.25 mm (trailing edge)
center of mass: (1.22 ± 0.01) mm behind leading edge
special feature: weighted front, sinusoidal trailing edge
Reynolds number∗ : Re ≈ 7.5 · 104

Figure 1: Picture of the X-Zylo

The object flies in an almost straight line and seems to lose very little height while airborne
(see picture 2). With the X-Zylo being simply a thin hollow cylinder as seen in figure 1, this
flight characteristic is quite impressive. To understand the flight behavior in great detail, a six
degree of freedom model was developed to compute the trajectory of an X-Zylo. To be able to
accurately calculate the aerodynamic forces and moments acting on the object during flight, CFD
simulations were conducted on a computing cluster. In order to test the prediction systematically,
a launch device with the ability to launch the X-Zylo in a controlled manner was developed
and build. The flight of the object is tracked using several cameras to obtain detailed flight
information; this data is then corrected for camera induced errors. In the end the theoretical
predictions are extensively compared to the observed data.
Many papers already simulated the flight of a spin-stabilized disc, widely known as Frisbee™,
which shows a quite similar flight behavior [2, 3]. However, a less rigorous approach on the
experimental part mostly defies a good comparison between theoretical and experimental trajectory.
This work aims to resolve this issue by using a dedicated launch mechanism for better control on
the initial flight parameters. The X-Zylo itself was also subject of former investigations [4, 5],
but in less depth than in the present study. Future applications could involve the optimization
of such toys as well as potential insights in annular airfoil technology experimentally used for
coleopters in the past. Furthermore the understanding of the aerodynamics of such elementary
objects could yield insight into the flow past more complex structures.
In the following work those thin hollow cylinders investigated are referred to as throwing rings
or simply rings. The “X-Zylo” is hereby only a particular, commercially available ring with the
special features stated above, which was used for all experiments conducted.

observed trajectory

Figure 2: Typical trajectory for an X-Zylo launched with a small launch angle. Even though
the ring flies more than 40 m, the flight path is extraordinary flat.
∗
reference values: v = 17.35 m/s, characteristic length is the ring’s chord length lavg

Page 1/37
Theoretical and Experimental Investigation into the flight of an X-Zylo Nils Wagner

2 Theory of Flight
Two technical terms have to be differentiated, as they are of great importance for the initial
flight and can be misunderstood easily. At first there is a specific launch angle, which is the
angle between the velocity vector at launch and its projection onto the xy-plane (note the used
coordinate system in figure 5). It captures how steep the X-Zylo is thrown in respect to the
ground. The second important angle is the Angle of Attack (AoA). The AoA is the angle between
the ring’s symmetry axis and the velocity vector. It has to be emphasized that the initial AoA
at launch and the launch angle are independent of each other; one can throw the ring very flat,
however with its axis inclined to give it a high initial AoA.
As stated before, the characteristic of the X-Zylo is the stable and straight flight. This holds
for every initial launch angle. The ring typically is thrown without an initial AoA since the
axis of the ring matches the direction, in which the ring is launched at start. This is especially
true for the launch system built, see section 6.4. This begs the question why the ring does not
fall to the ground as any other object would, considering that the ring is virtually rotationally
symmetrical, therefore generating no lifting force. The sinusoidal tracing edge of the X-Zylo
breaks this symmetry, but even without this wavy edge the ring will fly nonetheless. For simplicity
only idealized hollow cylinders with a straight trailing edge will be discussed in the following.
Also the air is seen as stationary, therefore the ideal scenario is windless.

2.1 First Drop and Equilibrium Phase

At launch it holds true that the ring has an AoA of 0°, meaning the symmetry axis as well as the
flight direction are parallel (see figure 3a). As the ring generates no lift force, it gets accelerated
towards the ground due to gravity. This results in a direction change of the velocity vector, while
the gyroscopic stabilization—due to the rapid spin imparted at launch—keeps the axis direction
of the ring (nearly) constant. Therefore, even after a short amount of time an increasing AoA
between the ring’s flight direction as well as ring’s axis will form shown in figure see figure 3b.
This imparts a linearly increasing lift force until the AoA is great enough to support the weight of
the ring. This initial flight phase is further denoted as the first drop. At a specific angle this lift
force compensates the gravitational pull and the ring encounters an equilibrium phase, where the
AoA is stable. If the lift force exceeds the gravitational force, the ring gets accelerated upwards
and therefore the AoA decreases, decreasing lift; and vice versa. Hence the ring flies straight for
an elongated time since this equilibrium state is maintained. Drag decreases the velocity of the
ring, and will slowly increase the equilibrium AoA since a higher AoA is needed to generate the
lift force equivalent to gravity. Ever-increasing AoA yields flow separation later during flight so
that the ring plummets quickly.

F~lift

F~drag ~0 F~drag α ~ kL
~0
~v0 k L L

z ~v
y
F~grav ω0 F~grav ω
x
(a) Initial launch condition (idealized, αlaunch = 0°). (b) Ring mid-flight (equilibrium).

Figure 3: Simplistic model of the flight of a throwing ring without consideration of torques acting.
While the idealized initial conditions show no angle between velocity vector and ring axis, during
flight an AoA α is formed due to the gravitational acceleration.

Page 2/37
Theoretical and Experimental Investigation into the flight of an X-Zylo Nils Wagner

It has to be noted that when thrown by a human, the initial condition with 0° AoA is not
perfectly met, additionally the ring wobbles a lot and stabilizes only after a short amount of
time [4]. This makes the flight hard to predict since the launch conditions are hard to measure
accurately. The influence of an initial AoA is further investigated in section 6.4.
Even though the flight can be explained well using those descriptions, which were formerly
known (see [6]), it can be observed that the ring drifts sideways later during flight. This effect
will be explained further in the following section since it was mentioned in former publications
(e.g. Tarr [5]) but not analyzed in detail.

2.2 Late Flight and Second Drop

Later during flight torques on the flying ring, that impact the direction in which the ring flies,
become more important. There are several key components that decide how much the direction
of the ring changes in the sideways y-direction:

i) Position of the Center of Pressure: As Center of Mass (COM) and Center of Pressure
(COP) do not fall together in a single point, a torque is imparted which lets the ring precess.
For traditional airfoils it is known that the COP moves upstream for higher AoA [7, pp.
385+386], which complicates the torque calculation. The CFD results in section 3.3.5 also
confirm that the COP is not fixed for the X-Zylo during flight (see figure 10). As the COP is
usually found to be at approximately a quarter of the chord length for a flat plate (quarter
chord point), the X-Zylo is designed to counter this by having a weighted front using a
thin metal band. This shifts the COM towards the quarter chord point so that the acting
torques are of small magnitude.

ii) Angular Velocity: During launch the ring spins rapidly, but friction decreases this spin
midst flight, so that the translatoric as well as the rotational speed decreases while airborne.
This results in torques becoming more prominent later on since the angular momentum of
the ring decreases over time.

iii) Aerodynamic Forces: Since the only forces which produce a net torque on the rotating
cylinder are aerodynamic forces, the torque is directly proportional to the magnitude of lift
and drag. As those forces are very small at launch due to the small AoA, the change in
angular momentum is not visible. However, the AoA increases steadily over time magnifying
aerodynamic forces. This is coupled to the decreasing translatoric velocity of the ring which
in contrast decreases lift and drag.

As the aerodynamic forces act in the xz-plane at launch, the torque will purely act in the
y-direction at first (see figure 4a). This torque slowly lets the ring precess, turning the ring
sideways. However the velocity vector remains unchanged at first, only the ring precesses (see
figure 4b). The tilt of the symmetry axis towards the flight direction and therefore towards the
oncoming air generates a sideways lift force. Only then does the velocity vector follow the ring
axis vector, letting the ring drift sideways, which can be observed. The direction in which the ring
drifts is therefore dependent on the spin direction imparted at launch along with the location of
the COP (behind or in front of the COM). It will become visible in section 6 that as the location
of the COP changes, also the drift direction changes midst flight.
A force not accounted for in this approach is the Magnus force acting on the spinning cylinder
when swerving sideways, effectively creating a sideways incident flow component. This component
however is negligibly small, only when dealing with stronger sideways winds, those forces have to
be considered. In the ideal windless case, this factor can therefore be neglected.

Page 3/37
Theoretical and Experimental Investigation into the flight of an X-Zylo Nils Wagner

6 0
Ly =
vy = 0
~ aero
M ~
L
F~lift
β
~ ∈ ~a ∈ R3 ∧ ay = 0

COP ~v , L ~v
F~drag
y F~aero COM
ω

x ω
z

(a) Initial cruising condition with torqe acting on the (b) Ring condition after torque changes
COM due to aerodynamic forces. the direction of angular momentum.

Figure 4: Simplistic model of how torques act on the ring during flight shown in birds-eye
perspective. The sideways drift angle β is exaggerated for better visibility. All vectors depicted
in the picture are projections on the xy-plane, the z-component is not shown.

3 Trajectory Calculation
To predict the flight behavior a program was developed which approximates the trajectory
numerically. A forward Euler method is chosen in which the solution is propagated using small
discrete time steps; the implementation was done in MATLAB [8]. The Euler method showed to
be sufficient for this problem, as a test using Adams-Bashforth methods of the second and third
order showed insignificant discrepancy between the results.

3.1 Basic Mechanics

Starting the calculation, some initial parameters have to be specified, for example the velocity
magnitude at launch vlaunch , the launch angle αlaunch , the launch height h(t0 ), and the angular
velocity ω(t0 ). Furthermore, the initial time is set to t0 = 0 and a discrete time step ∆t chosen for
the iterative forward Euler method. In addition, some ring parameters have to be specified, e.g.
mass m, inner radius of the hollow cylinder ri , outer radius ra , and length l. Assuming a uniform
mass distribution, the inertia of the hollow cylinder is then calculated to be I = 12 m ri2 + ra2 .

The position of the tip of the ring (point

on its axis in the plane of the leading edge) z
is named Ptip . All other used ring locations
are named using P with the point specified as
subscript. The normalized ring axis direction
vector is called R~ axis . For the launch conditions ~v (t0 )

one gets
h(t0 )
  αlaunch
vlaunch · cos(αlaunch )
~v (t0 ) =  0  , (1)
vlaunch · sin(αlaunch ) x
 
0 −y
P~tip (t0 ) =  0  , (2)
h(t0 )
Figure 5: Fixed coordinate system used
~ axis (t0 ) = ~v (t0 ) . throughout the work with a schematic tra-
R (3) jectory.
|~v (t0 )|

Page 4/37
Theoretical and Experimental Investigation into the flight of an X-Zylo Nils Wagner

In equation (3) the simplification that at start R~ axis is parallel to ~v (t0 ) is used. Note that this
is sufficiently true for the launch construction, in a human induced launch with its perturbations
this will not hold true (see section 6.4). With the given distance between the COM and the tip,
the position of the COM can be calculated by

P~COM (t0 ) = P~tip (t0 ) − P~COM (t0 ) − P~tip (t0 ) · R

~ axis (t0 ) . (4)

A loop variable n is introduced and set to 0 initially. As long as Ptip, z (tn ) > 0 holds, the ring is
considered airborne and equations (5) to (12) are repeatedly solved for the next iteration until
this condition is not satisfied any more.
All current forces acting on the ring are summed up and used to calculate the velocity and
positions for the next iteration. For the acceleration one gets

F~tot (tn ) F~lift (tn ) + F~drag (tn ) + F~grav (tn )

~atot (tn ) = = (5)
m m
with F~grav (tn ) = −mg · e~z using g = (9.81 ± 0.02) m/s2 . The velocity and position of the COM
for the next time step are then given by

~v (tn+1 ) = ~v (tn ) + ∆t · ~atot (tn ) , (6)

P~COM (tn+1 ) = P~COM (tn ) + ∆t · ~v (tn ) . (7)

Section 3.2 will focus on the magnitude of lift and drag as well as the COP; those quantities are
interpolated from the CFD results. The direction of the lift and drag vector can be calculated
using the Gram-Schmidt process resulting in
D E
~ axis (tn ) − ~v(tn )2 · R
R ~ axis (tn ), ~v (tn )
|~v (tn )|
F~lift (tn ) = E · F~lift α(tn ), |~v (tn )| , (8a)


D
~ axis (tn ) −
R ~v (tn )
2 · ~ axis (tn ), ~v (tn )
R
|~v (tn )|

~v (tn )
F~drag (tn ) = − · F~drag α(tn ), |~v (t0 )| , (8b)


|~v (tn )|

where h~a, ~bi denotes the standard scalar product in euclidean space of vectors ~a and ~b. The lift
and drag forces are dependent on the angle α(tn ) between the ring axis direction vector and the
velocity vector. One gets
D E
~ axis (tn ), ~v (tn )
R
α(tn ) = arccos  . (9)
~ axis (tn ) · |~v (tn )|
R

Note that as stated in section 2.2, the lift force does not always act in the xz-plane. After the
ring tilts sideways this will be accounted for in the direction of the aerodynamic forces (8a+8b)
as well as the AoA (9) used in the calculation of lift and drag.
~ axis changes over time. With
Due to torques acting on the ring the direction of the ring axis R
the angular momentum L(t ~ n ) being parallel to the axis, one deduces

~ n ) = L(t
L(t ~ n) · R
~ axis (tn ) = I · ω(tn ) · R
~ axis (tn ) . (10)

Of significant relevance is the position of the COP, which is evaluated using the interpolated
CFD results. A stated before, COM and COP do not fall together, therefore creating a torque
~ (tn ) acting on the ring:
M
   
~ (tn ) = P~COM (tn ) − P~COP (tn ) × F~lift (tn ) + F~drag (tn ) .
M (11)

Page 5/37
Theoretical and Experimental Investigation into the flight of an X-Zylo Nils Wagner

This torque changes the angular momentum so that one gets L(t ~ n+1 ) = L(t
~ n ) + ∆t · M
~ (tn ). The
new direction of the ring axis R~ axis (tn+1 ) is then the normalized angular momentum vector.
For the sake of simplicity the magnitude of the angular frequency is seen as decoupled from
the complicated motion of the X-Zylo. Using Sliding Mesh simulations (see section 3.3.6), the
mean Wall Shear Stress τw is calculated for different rotational frequencies and interpolated for
the trajectory simulation. Using only the angular velocity one can then calculate the additional
torque acting on the ring, which in turn reduces its angular momentum using the simplified
relation
~ n+1 ) = L(t~ n ) − τw ω(tn ) · Aring · q ω(tn ) · ra
(12)


L(t
2
ω(tn ) · ra + |~v (tn )|2
with the rings surface area Aring and the ratio of the rotational velocity ω(tn ) · ra to the total
velocity on the rings surface. Mind that equation (12) only holds for small AoA as the ratio of
rotational velocity and total velocity only holds for zero AoA flight. Therefore it is only a rough
approximation for larger AoA scenarios later in flight.
At last the loop variable n gets incremented by one, the time incremented by ∆t, and after
that all steps will repeat.

3.2 Aerodynamic Forces

As can be seen in the theoretical calculation in section 3.1, the magnitude of the lift and drag
force as well as the COP is needed for any arbitrary angle α(tn ) and any flow velocity |~v (tn )|.
As all magnitudes change gradually, the approach will be to calculate lift and drag coefficient
as well as the COP for discrete AoA using CFD (Computational Fluid Mechanics) and then to
interpolate the results. The change in Reynolds number and therefore the change in drag and
lift coefficient for different flow velocities is neglected. As the velocity magnitude of the ring lies
between 6 m/s and 18 m/s for the whole flight duration in most standard cases, this assumption
should yield fairly accurate results. Nonetheless, this is a potential source of error one has to
keep in mind, see section 6.5.

3.2.1 Theoretical Approximations

Several sources have derived analytical approximations for the lift and drag forces of hollow
cylinder configurations which will be shortly mentioned and later compared to the CFD results in
section 3.3.5. For clarity, the lift and drag coefficients (CD and CL ) are written in square brackets
as only the force equations are shown.
One of the first analytical descriptions was done by Ribner [9], who used concepts of Prandtl
lifting-line theory in order to derive the lift force for a thin ring airfoil of diameter d and (chord)
length l. This resulted in the expression
" # " #
2d
ρv 2 π2α ρv 2
Flift = πl
· π 2
α · · dl = · · dl , (13)
1 + 2dπl
2 1 + πλ
2
2
| {z } | {z }
CL [9] CL [10]

where λ = l/d is the aspect ratio between chord length and diameter of the ring, S = dl the
reference surface area for the coefficient calculation, v the velocity magnitude of the oncoming
air, and ρ the air density. Mind differences in the definition of the reference wing surface area S
in different publications, sometimes S 0 = πdl as well as S 00 = 2dl are used. Moreover, in other
literature the definition for the aspect ratio λ is sometimes defined as the reciprocal fraction.
Formula (13) was also found by Pivko [10], who additionally calculated the induced drag force to
be  !2 
λ 2
π α 2
Fdrag, induced =  ·  · ρv · dl (14)
2 πλ 2
1+ 2

Page 6/37
Theoretical and Experimental Investigation into the flight of an X-Zylo Nils Wagner

using the found correlation CD = λ/2 · CL 2 .

Weissinger [11] developed a refined theory on general ring wing configurations and found the
formulae (13) and (14) to be a special case for λ →− 0. A more accurate approximation for the lift
force for small aspect ratios (λ < 5) was found to be
" #
π2α ρv 2
Flift = · · dl . (15)
1 + πλ
2 + λ · arctan(1.2λ)
2

For rotationally symmetric rings the drag can be calculated analogical as in formula (14) using
CD = λ/2 · CL 2 [11].
Tarr derived another analytic expressions for lift and drag forces in his book “What Makes
The Amazing X-Zylo Fly” [5], specifically to examine the X-Zylo, yielding

ρv 2
Flift = [4πα] · · dl , (16a)
2
π · µ · v 3/2 · d · l1/2
Fdrag = 4π · ρ · v 2 · l2 · α2 + · (Supper + Slower ) (16b)
| {z } ν 1/2
induced drag
| {z }
viscous drag

where µ is the dynamic viscosity and ν the kinematic viscosity of air, Supper and Slower are the
velocity gradients on the upper and lower surface. Those gradients were numerically computed by
Tarr for an AoA of 1.32° and a flow velocity of 15 m/s, yielding Supper = 0.3172 and Slower = 0.3466.
The same values were used for the comparison seen in figure 10 while neglecting changes in the
velocity gradients based on the angle of attack α and the flow velocity v.
For later comparison with the CFD results, the coefficients were calculated for a temperature
of 20 °C, yielding ρ = 1.204 kg m−3 , µ = 1.825 · 10−5 kg m−1 s−1 and ν = 1.516 · 10−5 kg m−2 .

3.3 Computational Fluid Dynamics

To get reliable values for the aerodynamic forces as well as the COP, CFD simulations are
used. It is undoubted that todays CFD solvers (when used correctly) can outperform even the
best analytical models due to complex turbulence modeling and consideration of viscous effects.
Therefore even for higher AoA beyond flow separation, approximate results can be obtained.

3.3.1 Geometry and Mesh

The geometry of an X-Zylo, which is used for all simulations other than the validation cases, is
presented in detail in figure 6. All meshes are created in the same process. SALOME [12] is used
to create a structured quad surface mesh using quadrangle mapping, which can be seen in figure
7a for the X-Zylo. From there ANSYS Fluent [13] is used to generate the volume mesh (see figure
7c) as well as the CFD calculations itself. The boundary prism layer consists out of 20 layers
using a geometric layer height increase of 1.2. Hereby the initial height is set to satisfy y+ ≈ 0.5,
which for the X-Zylo results in an initial layer height of 0.01 mm resulting in a total boundary
layer height of approximately 1.87 mm (see figure 7b). A sphere of radius 2.5 m is used as far-field;
the volume mesh is an unstructured tetrahedral mesh. The simulations were conducted on the
Linux-Cluster of the LRZ (Leibnitz-RechenZentrum Garching, DE), utilizing the HPC resources
to cut computing times.

Page 7/37
Theoretical and Experimental Investigation into the flight of an X-Zylo Nils Wagner

lmax = 61.0 mm
lavg = 54.5 mm
outer side face leading edge
lmin = 48.0 mm
dfront, inner = 95.0 mm
douter = 97.0 mm
drear, inner = 96.5 mm

inner side face

sinusoidal
trailing edge

a) front view b) side view

inner side face 13.0 mm

35.0 − 48.0 mm
0.75 mm

outer side face c) profile 0.25 mm

Figure 6: Details of the geometry of an X-Zylo. The thickness is scaled by a factor of 4 in the
drawings for better visibility. The sinusoidal trailing edge consists of 5 full sine waves with
ampitude 13 mm. The profile drawing shows the lower part of the cross-section of the ring.

(a) Surface mesh (b) Boundary prism layers (c) Volume mesh

Figure 7: Mesh for the X-Zylo (M2), which showed small error from finer meshes as well as a
justifiable computing time.

3.3.2 Simulation Settings

The simulation settings were kept constant throughout the whole work, only specific setting that
change from case to case (e.g. the validation cases) are mentioned separately in the respective
paragraphs. For the simulation itself the transient pressure-based RANS (Reynolds-Averaged
Navier Stokes) solver is used. The Pressure-Velocity-Coupling was set to use the SIMPLE
(Semi-Implicit Method for Pressure-Linked Equations) solver, the relevant discretisation schemes
(pressure, density, momentum and energy) were set to second order. The ring itself was modeled

Page 8/37
Theoretical and Experimental Investigation into the flight of an X-Zylo Nils Wagner

to be a no slip wall while the sphere served as a pressure-far-field as inlet—therefore the air is
modeled as an ideal gas. To initialize the flow field, ANSYS Fluent’s standard initialization from
the pressure-far-field was used. As all simulations were conducted transient, the time-step was set
for the maximum CFL (Courant) number to be between 1 and 10, depending on the convergence
of the problem. All obtained values are for a temperature of 20 °C. As turbulence modeling is a
key component using the RANS solver, and many different models are implemented in ANSYS
Fluent, two validation cases were simulated using a variety of turbulence models.

3.3.3 Validation
There exist several sources that experimentally evaluate different forms of annular airfoils. Early
experiments done by Fletcher [14] were performed using annular airfoils with Clark-Y cross-section
and aspect ratios of 1/3, 2/3, 1, 3/2, and 3. Chord length and diameter varied from 0.235 m
to 0.704 m, the flow velocity used was 44.6 m/s. Therefore the dimensions of the used models
as well as the flow velocity greatly exceed the operating conditions of an X-Zylo with Reynolds
numbers 11 to 33 times greater than in the present case. A more recent paper by Traub [15]
examines annular wings with an Eppler-68 section and aspect ratios 1/2 and 1. Those wings were
merged into a NACA 0012 cross-section on the vertical sides allowing the Eppler-68 profile to be
normal at the bottom and the top section. Also a revolution of a NACA 0012 profile (λ = 1/2)
was tested by Traub [15, 16]. The chord length used for every model was 0.1 m, therefore the
dimensions resemble the studied case in this work better. The free-stream velocity used in the
wind tunnel was 40 m/s, which is still well above the flow velocities for the X-Zylo, yielding a
Reynolds number still 4.2 times greater than needed. Another experimental investigation was
done by Latoine [17] on small circular hollow cylinders with flat-plate cross-section. The largest
cylinder had a diameter of 6 in (≈ 0.152 m) and an aspect ratio of 1/8. Since the used flow
velocity was only 12.2 m/s, the Reynolds number based on the chord length is in this case 4 times
smaller than that of an X-Zylo in flight. Additionally, in this work the aspect ratio of λ = 1/8 is
far from the ratio of an X-Zylo with λ = 0.56 .
Due to the lack of experimental data in the exact Reynolds number regime of the X-Zylo,
both data sets by Traub (closed NACA0012 revolution) and Latoine were chosen to test the setup.
The Reynolds number of the studied case then lies between the numbers of both validation cases.

Experiment [17]
0.4 0.12
Experiment (adjusted)
laminar (0 eq.)
drag coefficient CD
lift coefficient CL

0.10 Spalart-Allmaras (1 eq.)

0.3
k−ω−SST (2 eq.)
Transition SST (4 eq.)
0.08 RSM (7 eq.)
0.2
Experiment [17]
laminar (0 eq.)
0.06
Spalart-Allmaras (1 eq.)
0.1
k−ω−SST (2 eq.)
Transition SST (4 eq.)
0.04
RSM (7 eq.)
0.0
0.0 2.5 5.0 7.5 10.0 12.5 15.0 17.5 −5.0 −2.5 0.0 2.5 5.0 7.5 10.0 12.5
angle of attack [◦] angle of attack [◦]

(a) Comparison of the lift coefficient (b) Comparison of the drag coefficient

Figure 8: Comparison of the CFD results to the experimental results obtained by Latoine [17].

In figure 8 the CFD results compared to the experimental data on the small hollow cylinder
by Latoine [17] are presented. The tested cylinder has a diameter of 0.152 m and a thickness of
0.51 mm; the mesh consisted of 3.85 million cells. Wind tunnel experiments were performed with a
wind speed of 12.2 m/s and a turbulence intensity of 0.02%; those parameters were also set in the
simulation. From there 2000 time steps with a step size of 5 · 10−4 s (CFL ≈ 10) were calculated.

Page 9/37
Theoretical and Experimental Investigation into the flight of an X-Zylo Nils Wagner

Several turbulence models, which have great influence on the CFD solution, are compared. As
the Reynolds number is very low, also a laminar solution is computed which fits the experimental
data well for very small AoA up to 2°. However, for larger AoA the solution diverges as small
flow separations form, especially when comparing the lift coefficient. All simulations involving
turbulence modeling can predict the large flow separation occurring at approximately 8° onward,
but overestimate the decrease in lift. The Transition SST (Shear Stress Transport) model hereby
is the best model while the Reynolds-Stress-Model (RSM) greatly underestimates the lift force
after separation. For smaller AoA this trend is reversed, RSM and k-ω-SST can predict the lift
slope accurately while Transition SST and Spalart-Allmaras (SA) compute a lift slope far greater
than captured in the experiment. In this low Reynolds regime the flow is not fully turbulent,
which is an explanation for the deviation between the models. All models can predict the drag
sufficiently well; except the RSM, which heavily over-predicts drag. The Transition SST model is
especially good at capturing the drag at small AoA. It has to be noted that the drag should be
fully axis-symmetric to an AoA of 0° since the hollow cylinder investigated is symmetric, however
the measured drag coefficient in the experiment showed a shift of approximately 0.8° and is not
symmetric. When adjusting this shift, the drag values fit the simulation data almost perfectly
(see figure 8b). From this validation case a laminar calculation as well as a simulation using the
RSM turbulence model can be excluded. Therefore, the second validation case was only simulated
using the remaining three turbulence models SA, k-ω-SST and Transition SST.
In the second validation case a closed NACA0012 revolution (λ = 1/2, chord length 0.1 m)
including its mount was simulated at wind tunnel conditions of 40 m/s wind speed and a turbulence
intensity of 0.24% [16]. 3000 time steps with a step size of 1 · 10−4 s (CFL ≈ 6) were calculated.
A finer mesh consisting of 8.21 million cells was used due to a more sophisticated geometry. The
results are displayed in figure 9. One can deduce that the setup works very well for low AoA where
little flow separation is present. The lift slope deviates from the experimentally measured one
by 8.9 % (SA), 10.2 % (k-ω-SST) and 5.4 % (Transition SST). Calculating the average deviation
of the drag coefficient up to an angle of 10° one gets errors of 4.7 % (SA), 4.6 % (k-ω-SST) and
15.9 % (Transition SST). For higher AoA all turbulence models struggle again due to large flow
separations.

1.2
0.36 Experiment [16]
Spalart-Allmaras (1 eq.)
0.9 0.30 k−ω−SST (2 eq.)
drag coefficient CD
lift coefficient CL

Transition SST (4 eq.)

0.6 0.24

0.18
0.3
Experiment [16]
0.12
0.0 Spalart-Allmaras (1 eq.)
k−ω−SST (2 eq.)
0.06
Transition SST (4 eq.)
−0.3
−5 0 5 10 15 20 −5 0 5 10 15 20
angle of attack [◦] angle of attack [◦]

(a) Comparison of the lift coefficient (b) Comparison of the drag coefficient

Figure 9: Comparison of the CFD results to the experimental results obtained by Traub [16].

Both validations show, that the simulation can qualitatively predict the flow even with large
flow separations occurring at higher AoA. From a quantitative standpoint the simulations are only
valid for small AoA which is the expected behavior. Nonetheless, both SA and the k − ω − SST
turbulence model produce good results in both validation cases. Transition SST showed good
performance in specific situations (low AoA drag for low Reynolds number case), but was not as
consistent. As the lift in the first validation case was best approximated using the k-ω-SST model,
it will be the choice for the simulations of the X-Zylo. It was later observed that for the case

Page 10/37
Theoretical and Experimental Investigation into the flight of an X-Zylo Nils Wagner

of the X-Zylo at small AoA, the Transition SST model predicts a drag coefficient 25% smaller
than that calculated with the k-ω-SST model. As the drag coefficient in the small Reynolds
number case was better approximated with Transition SST and the difference is significant, the
calculations were repeated using the Transition SST model to compare both results with the
found experimental behavior (see section 6).
As further validation a paper by Werle [18] summarized most of the available experimental
and CFD data to test the predictions by Weissinger [11]. It was shown that the lift slope was
almost exactly predictable for all cases using Weissingers approximation (15). Therefore also this
can be used as further validation for the obtained CFD results for the X-Zylo.

3.3.4 Mesh Independence Study

A small mesh independence study for the two SST turbulence models was done on four meshes
with varying sizes, ranging from 2 to 17 million mesh cells, roughly doubling every step. The flow
velocity in all further computations involving the X-Zylo was set to 17.35 m/s (Mach 0.05 at the
pressure far-field) and a turbulence level of 1% was used. Those values were estimated as the
standard environment of an X-Zylo flying in a sports hall. For every mesh a CFL number of 1
was used in the calculation and 0.5 s with an AoA of 5° were simulated. The results obtained
using the k-ω-SST model are visible in table 1, the results for the Transition SST calculations are
only summarized. It can be observed that especially drag and lift force can be well approximated
using even the coarsest mesh. In contrast the error for the COP— calculated from the finest
mesh (M4)—is still high using meshes M1 or M2. Since the COP was seen less important than
the aerodynamic forces, the coarse mesh (M2) was used in all further simulations as it was a
good balance between simulation time and the mesh induced error. Later it was discovered that
the COP is an extraordinary sensitive parameter and one should have opted for the fine mesh
(M3), see section 6.3. The error for forces calculated using the medium mesh (M2) lie well below
5 % with both turbulence models (0.3 % for k-ω-SST, 1.4 % for Transition SST), therefore the
influence of the turbulence model exaggerates the mesh induced error.

CD CL COP location
Mesh refinement number of cells % error
(% error) (% error) (% error)
0.08742 0.3714 22.99 %
coarse (M1) 2,268,018 3.69 %
(3.51 %) (0.57 %) (6.98 %)
0.08474 0.3702 22.24 %
medium (M2) 4,237,776 1.35 %
(0.33 %) (0.24 %) (3.49 %)
0.08404 0.3694 21.72 %
fine (M3) 8,389,523 0.53 %
(0.50 %) (0.03 %) (1.07 %)
very fine (M4) 16,960,108 0.08446 0.3693 21.49 % −

Table 1: Results of the mesh independence study for the k-ω-SST turbulence model. The error
is calculated from the finest mesh (M4), the total error is the arithmetic average of the single
deviations. The COP is given by the percentual location on chord. The same calculations were
performed using the Transition SST Turbulence Model, obtaining slightly higher total error
margins of 5.90 % (M1), 3.23 % (M2), and 1.16 % (M3). However, independent of the turbulence
model, the largest error is always seen for the calculation of the COP.

3.3.5 Results
The CFD simulations for the X-Zylo were carried out using 5500 time steps (each 4.4 · 10−5 s,
CFL 1.5) totaling a simulated flight time of about 0.24 s. The mean over the last 500 time steps
was calculated to average small numerical fluctuations as well as periodic effects. As noted before
in section 3.3.3, both SST models show a significant discrepancy in their drag behavior for small

Page 11/37
Theoretical and Experimental Investigation into the flight of an X-Zylo Nils Wagner

AoA. Since neither solution can be disregarded using the validation cases, all computations were
made for both turbulence models. While in both validation cases the drag estimates for small AoA
using the Transition SST model were conceivably smaller than k-ω-SST, in the small Reynolds
number regime of the case by Latoine, this estimate fits the data better. As the Transition SST
model is designed to operate well in the turbulence transition range where the X-Zylo mostly
operates, it is not unlikely that it performs well in this case. The final conclusion can be drawn
when the calculated trajectories using the CFD data are compared to the experimental results in
section 6.

CD (CFD, k−ω−SST) CD (CFD, Trans−SST) CD Weissinger [11] CD Tarr [5]

CL (CFD, k−ω−SST) CL (CFD, Trans−SST) CL Weissinger [11] CL Tarr [5]
COP (CFD, k−ω−SST) COP (CFD, Trans−SST) CL Ribner [9] ∆ CFD
1.2 7.50 0.6 30

difference CFD solutions [%]

difference CFD solutions [%]

1.0 6.25 0.5 25

drag coefficient CD
lift coefficient CL

0.8 5.00 0.4 20

0.6 3.75 0.3 15

0.4 2.50 0.2 10

0.2 1.25 0.1 5

0.0 0.00 0.0 0

0 5 10 15 20 25 0 5 10 15 20 25
◦ ◦
angle of attack α [ ] angle of attack α [ ]
40 10.0
1.4 50
lift coeff. CL , drag coeff. CD
difference CFD solutions [%]
COP location on chord [%]

COP location on chord [%]

1.2 45
35 7.5
1.0 40

0.8 35
30 5.0

0.6 30

25 2.5 0.4 25

0.2 20

20 0.0 0.0 15
0 5 10 15 20 25 0 15 30 45 60 75 90
◦ ◦
angle of attack α [ ] angle of attack α [ ]

Figure 10: Obtained CFD results using ANSYS Fluent [13] in comparison to several analytic
approximations. The dotted solid line graphs show the deviation of the Transition SST model to
the k-ω-SST model. All CFD data is interpolated using cubic splines to be used in the trajectory
calculation in section 6.

The results of the simulations as well as the analytic approximations from several sources can
be seen in figure 10. While the CFD results use a non-rotating X-Zylo with wavy trailing edge,
the other sources are calculated for a thin ring wing with length 54.5 mm and diameter 97 mm.
As Hirata, et al. [4] calculated the drag and lift coefficient of a simplified model with a length of
60 mm and a diameter of 100 mm using motion analysis, the values are scaled accordingly to fit
the real dimensions of an X-Zylo. It also has to be mentioned that source [4] only lists a single
value for CD and CL at an AoA of 2°. Therefore this data is not listed in figure 10.
At first, it can be seen that while both SST turbulence models have a different drag behavior
at small AoA, they coincide well for higher AoA. The error for the lift force only exceeds 3% for
AoA smaller than 2°; the error for drag also plummets to under 10% for AoA greater than 4°.

Page 12/37
Theoretical and Experimental Investigation into the flight of an X-Zylo Nils Wagner

For the COP location also a small error of under 4% for AoA greater than 1° is seen. Altogether,
the error gets smaller the higher the AoA gets with the exception being the lift coefficient. Here
both turbulence models differ in the prediction of the flow separation, however not significantly.
It can also be noted that the X-Zylo seems to behave like a traditional biplane wing with its lift
curve showing two different slopes before and after large flow separation occurs.
As can be seen the approach by Tarr [5]—equations (16a) and (16b)—greatly overestimates
the lift and induced drag force generated by the X-Zylo, while underestimating viscous drag. For
the lift slope the deviation to the CFD results is more than 180%, also the viscous drag is only
half of what was computed (42% for k-ω-SST, 56% for Transition SST). The approach by Hirata,
et al. [4] also underestimates drag and lift forces. When comparing the sole data point given to
the Fluent simulation a difference of 40% for the lift force (both models) and 41% (k-ω-SST) or
27% (Transition SST) difference for the drag force is calculated. The analytical approximations of
the lift force found by Ribner [9], Pivko [10] and especially Weissinger [11] coincide well with the
simulation data obtained. The difference in lift slope of the refined formula (15) by Weissinger
and the CFD simulation is less than 1% for both SST models which is remarkable. This also
serves as additional validation since formula (15) was found to be very precise before by Werle [18].
Nonetheless, only Tarr approximates viscous drag of the X-Zylo, therefore the drag curve for the
Weissinger approximation starts in the origin. The induced drag force calculated using formula
(14) with the correction term for CL by Weissinger also matches the induced drag calculated
using CFD for an AoA smaller than 5°, but diverges from the obtained simulation results for
angles greater than 5°.

Figure 11: Pathlines around a non-rotating X-Zylo showing the turbulent intensity along their
path for different AoA. The re-circulation zones are clearly visible with their size increasing for
rising AoA.

Figure 11 shows pathlines colored by turbulence intensity around the X-Zylo for high AoA to
qualitatively capture the flow separation. It can be seen that as the flow slowly separates in the
center of the upper and lower wing segment, the side flow swirls into this area due to the lower
pressure region created. For AoA smaller than 10° this effect is small and the separation bubble
is only concentrated to the center portion of the wing while growing outwards for higher AoA. As
the study of the flow separation is not a focus in this paper, this will not be investigated further.

3.3.6 Influence of the Model’s Rotation

To account for the rotation of the X-Zylo during flight, a Sliding Mesh approach was used to simu-
late the ring with different rotational frequencies. The simulations were conducted for an AoA of 5°
using the k-ω-SST model and a mesh containing 5.6 million cells, the results can be seen in figure 12.
The percentual difference between the non-rotating solution and the simulation with non-vanishing

Page 13/37
Theoretical and Experimental Investigation into the flight of an X-Zylo Nils Wagner

revolutions per second [Hz]

rotational frequency is shown. Especially 0 15 30 45 60 75
the influence of the rotation towards the

s solution [%]
15 CD deviation
drag is not negligible as the rotational fre- CL deviation
quency of an X-Zylo shot by the launch 12 COP deviation
CD interpolation
mechanism can exceed 50 revolutions per
9 CL interpolation
second (see table 2). It is seen that the

deviation from 0 rad

COP interpolation
increase in drag stems from a longer dis- 6
tance traveled by the air over the chord as
the air is deflected near the surface by the 3

rotating ring. This also increases lift and

shifts the COP, however this influence is 0

relatively small. The interpolation for CD 0 100 200 300

rotational frequency [ rad
400
]
500

and CL was done using a cubic polynomial,

s

the COP fitting only uses a quadratic poly- Figure 12: Results for a rotating X-Zylo at an
nomial. Those interpolations are then used AoA of 5° using a Sliding Mesh approach.
in the program to simulate the trajectory.
Moreover, the wall shear stress was captured for the rotating X-Zylo to calculate the decrease in
angular velocity over time for different rotational frequencies (see equation (12)).

4 Launch Construction

drilling machine

pivotable pulley

rubber roller

trigger spring
V-belt

mount + slide
central
guiding rod stopper
X-Zylo

launch slide
rubber bands
counterweight

base plate
side arm

Figure 13: Schematic of the launch construction. The different parts are labeled as well as the
actual object for launch marked in red. The functionality of the parts is covered in the text.

To be able to throw the ring in a reproducible and controlled fashion, a launch device was
constructed, which can be seen in picture 14. Especially the swerving motion of the ring at launch
when thrown by a human as well as a non-vanishing initial AoA (see section 6.4) complicate the
early flight behavior.

Page 14/37
Theoretical and Experimental Investigation into the flight of an X-Zylo Nils Wagner

4.1 Explanation of the used Mechanism

Figure 14: Picture of the cocked launch device which was developed and constructed for this
project. The X-Zylo colored in red sits on a mount which is then accelerated forward by rub-
ber bands. The rotation is achieved using a drill and a V-belt.

To explain the mechanism in a nutshell, schematic 13 shows the most important parts. The ring
sits loosely on a mount which can be altered for different ring geometries. This mount can slide
on a central guiding rod which is oriented in the direction the ring should be launched. The
initial launch angle can be chosen between 0° and 25°. The whole mount will be spun by a V-belt
connected to a drilling machine, a counterweight maintains tension in the belt. When the mount
spins at the desired speed, the weight is lifted, releasing the tension in the belt. The trigger can
be pulled to release the cocked launch slide, catapulting the launch slide as well as the mount
forwards using rubber bands (Thera-Band Gold). A stopper at the end of the guiding rod stops
the mount spontaneously and releases the X-Zylo; a spring dampens the hit on the stopper. To
evaluate the velocity and angular frequency of the ring at launch, a slow motion video of the ring
is captured (see section 5.2). A scale in the xz-plane is set up to calculate the initial velocity from
the video footage. When fully cocked, the launch mechanism has a draw force of approximately
1100 N − 1600 N, depending on the number of rubber bands used on each side arm.

4.2 Reproducibility of the Device

The reproducibility of the launch construction was tested by launching the X-Zylo five times
using the same settings and comparing the initial launch values (table 2) as well as the observed
trajectories (figure 15). Full reproducibility would be achieved if the standard deviation of the
launch parameters would be smaller than the estimated uncertainty on each individual parameter

Page 15/37
−10

Theoretical and Experimental Investigation into the flight of an X-Zylo Nils Wagner
−20

z coordinate [m]
launch 1
−30 launch 2
launch 3
launch 4
z coordinate [m]

launch 5
−40
x coordinate [m]

y coordinate [m]
−50

y co
ord ]
−60 ina
−100 −90 −80 −70 te−60
[m] −50 −40 −30 −20
x dinate [m
co or−10
x coordinate [m]

Figure 15: Comparison of the trajectory of five launches with the launch contraption. To test
the reproducibility of the device all launches were done using the same settings. The positions
shown are the COM locations.

evaluated using the camera setup (see section 5.2). This would then cut the uncertainty on initial
parameters due to the reliability of the construction.
From the trajectories it is visible that the goal of achieving reproducibility with the launch
mechanism was not met. While the flight distance varies only slightly, especially the sideways
drift is severely different for the launches. When comparing the data shown in table 2, it becomes
obvious that the disagreement in sideways drift stems from the huge difference in the initial
rotational frequency. This problem can be traced back to the belted motor drive mechanism. The
counterweight has to be lifted manually in order to lower the drill, releasing the tension of the
V-belt. In the same instance the trigger has to be pulled. However, this is hardly possible as
the trigger mechanism is poorly operable. Therefore, the tension in the V-belt is often released
before the trigger can be pulled, decreasing the rotation of the mount rapidly. Table 2 shows that
good reproducibility is almost achieved for the launch velocity, the launch angle still shows a high
standard deviation due to the outlier launch 3. However, the difference in the launch angle for

initial velocity launch angle init. rot. frequency launch time

vlaunch [m/s] αlaunch [◦ ] ω(t0 )/(2π) [Hz] distance [m] aloft [s]
launch 1 15.9 ± 0.2 11.8 ± 0.3 18.5 ± 0.4 38.9 ± 0.2 4.33 ± 0.07
launch 2 16.6 ± 0.4 12.3 ± 0.5 38 ± 2 39.7 ± 0.3 4.47 ± 0.07
launch 3 15.8 ± 0.3 10.2 ± 0.4 16 ± 1 38.6 ± 0.3 4.27 ± 0.07
launch 4 15.4 ± 0.4 12.6 ± 0.5 63 ± 3 37.0 ± 0.3 4.43 ± 0.07
launch 5 16.1 ± 0.2 13.0 ± 0.4 60 ± 3 38.3 ± 0.3 4.47 ± 0.07
median 16.0 ± 0.4 12.0 ± 1.1 39 ± 22 38.5 ± 1.0 4.39 ± 0.09

Table 2: Comparison of five launches using the launch contraption with the same settings. The
initial values for the velocity, the launch angle and the rotational frequency were determined
with uncertainty. The median of all five launches as well as the standard deviation is given for
comparison with the uncertainty estimated for each observation. Note that as launch 1 was
studied in detail (see section 6.1), the initial parameters were determined with greater precision.

the other four launches is small enough to be satisfactory. All in all the reproducibility fails due
to the rotational velocity (standard deviation more than ten times the individual uncertainty),
where the V-belt mechanism has to be improved. Nevertheless, the contraption is still more
reliable compared to a human induced launch and moreover satisfies the constraint of an initial
AoA close to 0°. This will prove necessary later, see section 6.4.

Page 16/37
Theoretical and Experimental Investigation into the flight of an X-Zylo Nils Wagner

5 Experimental Procedure and Trajectory Evaluation

5.1 Preparatory Measurements
Before conducting the experiments, the mass of the X-Zylo was measured using a high-resolution
balance. Measuring two traditional X-Zylo’s, an average mass of (22.73 ± 0.16) g was found.
Another X-Zylo was carefully disassembled and the plastic and metal part measured independently.
With the mass of the metal ring being (16.42 ± 0.01) g and the plastic weighing (6.27 ± 0.01) g
one can deduce the COM being (12.2 ± 0.1) mm behind the leading edge. The glue mass between
the metal ring and the plastic hull was negligible. Comparing the results to the measurements
done by Tarr [5] shows good agreement in the X-Zylo’s mass. A larger discrepancy for the COM
location is seen, which was found to be “at an axial distance of 1.3 cm from the leading edge” [5].
For the experiments the X-Zylo was colored red for better visibility and a black line was added
to calculate the rotational frequency from the slow motion footage. It was however found in
earlier tests that the additional mass from the red paint was not negligible and shifted the COM.
By weighing the painted X-Zylo it was found that the mass increased by 1.24 g (10.5%) and the
COM shifted from 12.2 mm to 13.0 mm behind the leading edge which significantly changed the
observed drifting behavior (see section 6.3).

5.2 Camera Setup and Corrections

The trajectory of the ring in the xz-plane was captured using a GoPro HERO 8 Black in linear
mode (4k, 60fps). Linear mode uses the dedicated GoPro-intern software to eliminate the barrel
distortion typically encountered with their cameras. Therefore only insignificant warp effects
are still visible which interfere with a qualitative analysis of the flight. Additionally, a Samsung
Galaxy S10 and S9 are both used as close-up high-speed cameras capturing 960fps at a resolution
of 720p. Another Samsung Galaxy S7 (4k, 30fps) is used to capture the sideways drift of the
X-Zylo during flight in the yz-plane. Finally, a GoPro HERO 4 Silver (linear mode, 1080p, 30fps)
was used as a backup camera to probe different locations during flight. The camera setup with
their positions can be found in figure 16. From now on almost exclusively the abbreviations S10,
S9, S7, GoPro8 and GoPro4 are used instead of the full camera names mentioned above.
3.0 m
42.0 m

Positions in the gym 4.7 m 5.0 m

3.0 m
calibration points 7 1 6
3.2 m
1 launch construction 2 6.8 m

2 Samsung Galaxy S9 (SloMo) 3

3 Samsung Galaxy S10 (SloMo) 28.0 m
2.8 m 22.8 m 22.5 m
4 GoPro Hero 8 Black
5 GoPro Hero 4 Silver (Day 1)
6 GoPro Hero 4 Silver (Day 2) 34.8 m
21.6 m
7 Samsung Galaxy S7 5
4
45.0 m

Figure 16: Sketch (to scale) of the camera and calibration setup in the school gym for the final
experiments. All camera positions shown were fixed, only the Samsung Galaxy S9 was moved
throughout the experiments.

Using the open-source software Tracker [19], the trajectories could be manually evaluated
frame by frame. The experiments were conducted in a school gym to reduce external factors,

Page 17/37
 Theoretical and Experimental Investigation into the flight of an X-Zylo Nils Wagner

former outside tests revealed bad flight behavior due to even slightly windy conditions. To
quantitatively capture the trajectory, both the xz- and yz-plane were equipped with calibration
points marking different distances. Additionally, a two-meter-long colored calibration bar was
used for the slow motion footage to accurately compute the launch velocity (see figure 19).
However, just observing the trajectory is not enough. Several effects have to be accounted for
to correct different imaging errors. As camera distortions are hard to deal with and are thought
to have negligible impact since they are already internally corrected for the GoPro footage, they
are ignored. Only purely geometrical corrections independent of the camera are discussed in the
following.

Ring outside the Plane of Measurement

The plane of measurement is here defined to be the plane perpendicular to the camera’s view,
in which the calibration points are located. Therefore the camera only captures the projection
of the ring onto this plane of measurement. If the ring is not located in this plane, one has to
account for this via the intercept theorem, which is demonstrated in figure 17.

xcamera ∆x real trajectory

xcorrect

xobserved
measurement plane
launch origin ∆d

y
dcamera
real X-Zylo positons
observed x positions
correct x positions
x camera
Figure 17: Using the intercept theorem and equation (17) one can correct the observed X-Zylo
positions. This is necessary as the X-Zylo leaves the measurement plane and therefore only a
projection is measured, which is illustrated in this figure.

As the ring swerves during flight, it leaves the xz-plane captured by the GoPro8. This results
in a correction term (see equation (17)) which has to be applied to the data. Mind that both ∆x
and ∆d are signed quantities. In figure 17 ∆d is negative and ∆x positive.
∆d
xcorrect = xobserved − ∆x = xobserved + (xobserved − xcamera ) · . (17)
dcamera
As the S7 films the sideways drift while the GoPro8 is capturing the xz-projection, both cameras
together can recreate the trajectory corrected for the swerving motion. The S7 however is also
subject to the intercept theorem since the calibration points are set up at the opposite end of the
1.00 gym. Therefore
1.02 the image
1.04 corrections of both cameras are coupled. The rectifications are applied
sequentially; first the S7 footage is adjusted with the x-coordinate of the GoPro8 video, then the
GoPro’s footage is corrected with the swerving motion of the S7. One could apply the corrections
again, however those higher order terms are of negligible magnitude.

Page 18/37
 Theoretical and Experimental Investigation into the flight of an X-Zylo Nils Wagner

−∆d+ra −∆d−ra
~vcorrect = ~vobserved · dcamera
dcamera ~vcorrect = ~vobserved · dcamera
dcamera

measurement plane

∆d
~vcorrect flight path
ra

dcamera
y
real X-Zylo positons
observed x positions
correct x positions
x camera
Figure 18: The intercept theorem also applies if the plane of measurement is not exactly located in
the flight path of the X-Zylo. If the camera is very close to the flight path, one has an additional
effect of observing different parts of the ring during flight.

Ring in close Proximity to the Camera

From the S7 footage it was seen that the X-Zylo trajectory was offset from the measurement
stick set up (see figure 20). This results again in an application of the intercept theorem for the
close-up cameras, illustrated in figure 18. However, this was not the only effect observed. When
looking closely at picture 19, which shows a time series of the tracked X-Zylo seen from the S10’s
slow motion footage, one spots that the X-Zylo is tracked at different locations throughout the
flight. As the object is probed almost precisely at its foremost point every frame, the spot tracked
with this method changes. This is also illustrated in figure 18. Before passing the camera, the
1.00 side facing
1.02the measurement
1.04 plane is tracked while after that the opposite side facing the camera
is tracked. As only the velocities in the slow motion footage were important, a simple application
of the intercept theorem results in the corrected velocities. This correction is important even if
the difference between the tracked points is only the diameter of the ring, therefore circa 5 cm.
This comes due to the close proximity of the camera to the flight path. As for example the S9 is
located only about 3.3 m from the measurement plane, one gets an offset of approximately 1.5%
which—for standard launch conditions with a velocity magnitude of 16 m/s—results in an error
of 0.24 m/s, that can not be neglected.

6 Comparison between Theory and Experiment

To compare the experimental results to the predictions made by the theoretical model, one launch
is studied in great detail. Furthermore, the influence of the launch angle is investigated and
compared to the model. At last, the sideways drifting behavior as well as the impact a human
induced launch causes are studied.

6.1 Detailed Results for a single Launch

Launch 1 seen in section 4.2 is used for the detailed study of a single launch. This is useful as
the initial rotational frequency is sufficiently low to observe the drifting behavior in detail. In
addition, the launch angle is optimal to achieve a long distance shot while also gaining enough
height for greater insight into the vertical components. Figure 19 shows a time series of this
launch observed by the S10 camera, while figures 20 and 21 present the trajectory of the X-Zylo.

Page 19/37
Theoretical and Experimental Investigation into the flight of an X-Zylo Nils Wagner

tip position every 10th frame

tip position every frame

αlaunch = 11.74◦ ± 0.58◦

(2.00 ± 0.02) m

(5.00 ± 0.10) m

Figure 19: Time Series of an X-Zylo immediately after launch. Every frame the position is
manually evaluated using Tracker [19], which are the small black points seen in the figure. The
larger red points show the tracked X-Zylo position every 10th frame. The uncertainty on the
calibration scales stems from being in slightly different measurement planes (see figure 20) as well
as an estimated scaling error due to camera distortions (see section 9.2).

The calibration points—as already seen in figure 16—are also marked in the pictures. Using
the scale uncertainty due to camera distortions, the lengths are afflicted with an error using the
values from table 4. Mind that the trajectories seen show only the raw footage and are therefore
not adjusted to the errors mentioned in section 5.2.
To get the initial launch parameters one can view the footage of the different cameras. As the
parameters change rapidly after launch, only the first five frames of each video are used. From those
frames the median as well as the standard deviation for the different values are computed using the
Student’s t-distribution assuming statistical scattering
observed trajectory of the values. Additionally, a systematic uncertainty is
calibration points
estimated for each camera. This yields
(3.00 ± 0.05) m
vlaunch = (16.15 ± 0.03 ± 0.30) m/s (S9) ,
= (15.84 ± 0.15 ± 0.15) m/s (S10) ,
= (15.5 ± 0.5 }) m/s
0.2 } |± {z
| {z (GoPro8) .
stat sys

(5.00 ± 0.08) m
Using those values the weighted mean can be calculated,
resulting in a launch velocity of (15.94 ± 0.21) m/s. No
difference between systematic and statistical error is
made after applying the weighted mean. Repeating the
procedure for the launch angle one gets
z
αlaunch = (11.86 ± 0.29 ± 0.20) ◦ (S9) ,
◦
= (11.74 ± 0.43 ± 0.15) (S10) ,
y ◦
= (11.6 ± 0.3 ± 0.5 ) (GoPro8) ,
Figure 20: Examined launch in
the xy-plane (raw footage of S7). which then results in αlaunch = (11.77 ± 0.34) ◦ . The
initial rotational frequency after applying the weighted
average is ω(t0 )/(2π) = (18.5 ± 0.4) Hz. Additionally, more insignificant uncertainties are set

Page 20/37
 Theoretical and Experimental Investigation into the flight of an X-Zylo Nils Wagner

S7
z
(3.00 ± 0.03) m
(5.00 ± 0.05) m

S9 (40.0 ± 0.4) m observed trajectory

S10 calibration points

x
Figure 21: Examined launch in the xz-plane (raw footage of GoPro8). The close up cameras are
marked as well as the calibration distances shown with uncertainty.

for the ring mass, the location of the COM, the launch height, the gravitational constant and
the ring thickness. Using those uncertainties one can calculate the theoretical trajectory with
uncertainty as described in section 9.1. However, the error margins for the CFD simulations also
have to be specified, e.g. using the mesh study (see table 1). The uncertainties for the Transition
SST model were set to be ±2% for CD and CL , and ±8% for the location of the COP; the values
for k-ω-SST are set to ±1% and ±5% respectively. For both turbulence models the uncertainty
for the mean wall shear stress τw was set to ±15% as equation (12) is only approximately correct.
experimentally
experimentally measured
measured values
values
initial values
initial
initial values
values experimentally
velocity:
velocity:
velocity:
measured
---15.94
--- ---m/s
15.94
values
m/s frequency:
frequency: ---18.5
--- Hz
frequency:
18.5 Hz
---
X-ZyloComparison
X-Zylo
X-Zylo Trans-SST
Trans-SST angle:
launchlaunch velocity [m/s]
velocity [m/s] 17.515.94+/-+/-
+/-0.30.21 angle: ---11.77°
angle:--- ---
11.77° X-Zylo
ring
X-Zylo
ring used:
used: ---X-Zylo
X-Zylo
ring --- ---
used:
X-Zylo
launch velocity [m/s] 15.94 0.21
launchlaunch
launchangle angle [°] [°]
angle [°] 911.77+/-+/-
11.77 +/-0.40.34
0.34
ringmass
ringring
mass mass [g] 22.73
[g] [g] 23.97+/-+/-
23.97 +/-0.01
0.01
0.01
center
center
center ofofmass
of mass mass [cm] [cm] 1.3
[cm] 1.30+/-+/-
1.30 +/-0.01
0.01
0.01
spin
spinspin frequency [Hz]
frequency
frequency [Hz]
[Hz] 60 18.5+/-+/-
18.5 +/- 30.4
0.4
ringlength
ringring
lengthlength [cm] [cm] 5.45
[cm] 5.45+/-+/-
5.45 +/- 0 - -
ringinner
innerradius
radius [cm] [cm] 4.75 +/-+/- - -
ringring
inner radius [cm] 4.75
4.75 +/- 0
ringthickness
thickness [mm] 1.00 +/- 0.05 0.05
ringring
thickness [mm][mm] 11.00 +/-+/- 0.05
launch
launch heigthheigth [m]
[m] 1.40 +/-
1.40 +/- 0.01 0.01
launch heigth [m] 1.4 +/- 0.01
timeinterval
time interval [s]
[s] 0.001
0.001
time interval [s] 0.001
xxaxis
axis: : centercenterofofmass
mass(x-coordinate)
(x-coordinate)
x axis : front of ring (x-coordinate)
yyaxis
axis: : centercenterofofmass
mass(y-coordinate)
(y-coordinate)
y axis : front of ring (x-coordinate)
zzaxis
axis: : centercenterofofmass
mass(z-coordinate)
(z-coordinate)
z axis : - - -
moresettings
more settings
more settings
aerodynamicforces
forces experimentaldata:
data:
aerodynamic experimental
aerodynamic forces
torquecalculation
torque calculation experimental data:
03.09.2020
03.09.2020
torque calculation
compare
compare parabola
parabola no comparison
launch11
launch
compare parabola
uncertainty calculation no comparison
uncertainty calculation showexp.
show exp.data
data
uncertainty calculation
equalaxis
equal axis show exp. data
calculate
calculate
equal axis
show
show legend
legend
calculate
show legend

Figure 22: 3D Trajectory of the X-Zylo in comparison with the theoretical predictions. The
interface seen here is an interactive MATLAB GUI written to easily probe the rings flight in
detail. Using the drop-down menus enables the user to plot almost all combinations of variables
calculated. For reasons of space from now on only the graph—then plotted in Python—is shown
without the GUI.

The theoretical trajectories for both turbulence models with their confidence interval are seen
in figure 22. The observed trajectory is plotted without uncertainty for better visibility. The
interactive GUI developed for the program shows the different initial values with uncertainty as
well as the different calculation settings. As this 3D view of the trajectory is only lightly helpful,
figure 23 shows the different quantities as 2D plots for better comparison.
Figure 23 presents the observed trajectory with derived quantities as well as the predictions
by both turbulence models. Additionally, for comparison a conventional flight parabola without
fluid forces is shown in several subplots. The uncertainty on the simulated quantities is calculated
using the uncertainty on the initial launch values and the procedure described in section 9.1. The
error margins of the observed values are calculated using the procedure in section 9.2, in which
the camera induced error is estimated. In the following every subplot of figure 23 is individually
evaluated.

Page 21/37
 Theoretical and Experimental Investigation into the flight of an X-Zylo Nils Wagner

simulation (Transition SST) observed values (S10, SloMo) observed values (GoPro 8) observed values (S7 + GoPro8)
simulation (k − ω − SST) observed values (S9, SloMo) observed values (GoPro 4) parabola comparison

4
a b c

y-coordinate (sideways drift) [m]

4 4
100 −90 −80 −70 3 −60 −50 −40 −30 −20 −10
z-coordinate (height) [m]

z-coordinate (height) [m]

trajectory
3 2 3

1
2 2

1 1
−1

0 −2 0
0 10 20 30 40 0 10 20 30 40 2 1 0 −1 −2
x-coordinate (distance) [m] x-coordinate (distance) [m] y-coordinate (sideways drift) [m]

16

d 3.0 e 16
f
14

v | [m/s]
14
1.5
x-velocity vx [m/s]

z-velocity vz [m/s]

12

velocity magnitude |~
12
10 0.0

10
8
−1.5

6 8

−3.0
4 6

2 −4.5
0 1 2 3 4 5 0 1 2 3 4 5 0 1 2 3 4 5
time [s] time [s] time [s]
16.00 3.50 16.25
g h i

velocity profiles
15.75 3.25 16.00
v | [m/s]
15.75
15.50 3.00
x-velocity vx [m/s]

z-velocity vz [m/s]

velocity magnitude |~

15.50
15.25 2.75
15.25
15.00 2.50
15.00
14.75 2.25
14.75

14.50 2.00
14.50

14.25 1.75 14.25

14.00 1.50 14.00

0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35
time [s] time [s] time [s]
5
10
j 12 k l
4
0
11
velocity angle α~v [◦]

velocity angle α~v [◦]

y-velocity vy [m/s]

−10 3
10
−20
2
9
−30
1
8
−40
0
−50 7

−1
−60 6
0 1 2 3 4 5 0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0 1 2 3 4 5
time [s] time [s] time [s]

120 11 40
m n o
atot| [m/s2]

10
auxiliaury info

35
110
angular frequency ω [rad/s]

9
30
angle of attack α [◦]

100
acceleration magnitude |~

8
25
90 7
20
6
80 15
5

70 10
4

3 5
60

2 0
0 1 2 3 4 5 0 1 2 3 4 5 0 1 2 3 4 5
time [s] time [s] time [s]

Figure 23: Comparison of the theoretical quantities with the observed trajectory data for a single
launch. Plots a-c show the trajectory in the xz-, xy- and yz-plane while plots d-l show different
velocity profiles. In plots m-o additional information not probed for the real trajectory is visible.

Page 22/37
Theoretical and Experimental Investigation into the flight of an X-Zylo Nils Wagner

23a) The projection of the observed trajectory into the xz-plane is shown. As can be seen, after
all corrections are applied to the raw footage as described in section 5.2, both camera
perspectives (GoPro8 and GoPro4) coincide really well in the given confidence interval.
When comparing the experimental data with the theoretical predictions, it can be seen
that during early flight, a good match could be achieved. However, later during flight the
paths diverge outside the joint error margin. The decrease in height (second drop) is also
more rapid in the observation compared to both predictions. This is only lightly visible
with the given aspect ratio of the plot, when having equal axis this becomes more obvious.
Nonetheless, the overall shape of the trajectory is in good agreement with the theory. Mind
that the z-axis is heavily scaled, therefore the trajectory seems not as flat as it actually is.
Lastly when comparing both turbulence models, it is seen that k-ω-SST gives better
predictions since it predicts a higher drag value for low AoA. Therefore the distance covered
by the ring is smaller which better resembles reality. Also it is seen that even though both
models differ quite significantly for low AoA drag prediction, the influence is rather slim.
This comes due to the first drop, as small AoA are only present for a short amount of time,
see figure 23o.

23b) When looking at the projection of the trajectory in the xy-plane, it is visible that the
qualitative behavior of the X-Zylo can be modeled well. The X-Zylo first swerves to the
right, then turns mid-flight and rapidly swerves left at the end of flight, as seen in figure
20. This can be explained using the model described in section 2. As the COP moves
upstream, it passes the COM of the ring at an AoA of approximately 6.5°. Consequently,
the torque acting on the ring changes direction in mid-flight when the COM is passed by
the COP which results in a direction change of the swerving motion. This torque sign
change is further denoted as torque flip. It can be seen that the sideways drift is stronger
than predicted, however the ring hits the ground earlier in the experiment. Therefore, the
X-Zylo is not able to drift as far at the end of flight as in the simulation. The sideways
drift also influences the observation in the xz-plane, as all spatial directions are coupled.
An increase in y-velocity reduces the other velocity directions accordingly and is therefore
visible in the xz-projection.
The uncertainty on the theoretical prediction is huge in this case due to the high error
margin imposed on the COP location (±5% for k-ω-SST, ±8% for Transition SST). In
addition to the uncertainty for the mean wall shear stress—which describes the decrease in
rotational velocity—one gets this large uncertainty. It is also visible that the additional 3%
error on the Transition SST’s COP data make a big difference as the uncertainty almost
doubles due to nonlinear behavior. It is later seen that the location of COP and COM are
very sensitive flight parameters, see section 6.3.

23c) The yz-projection is less interesting on its own, but it completes the whole picture. For
better visibility only the uncertainty bounds are shown. One can see the behavior also
observed in the real world, seen in figure 20. Mind that the values shown in the plot are
corrected for perspective, therefore both figures do not look perfectly alike. The qualitative
behavior is predicted correctly, however another time it is seen that the swerving to the
right is more pronounced than expected. Using this representation it is also better visible
that the ring looses height faster compared to the simulation at the end of flight. Even
though the observed sideways drift is more rapid at the end of flight, the slope late flight
shows a steeper descend.

23d) As can be expected the x-velocity decreases over time due to drag acting on the ring.
Again the behavior seen in the observed trajectory is also seen in the theoretical prediction.
The disparity seen seems to stem from a difference in the launch velocity. The theoretical
prediction shows an almost constant offset from the observed velocity, indicating that the
used launch velocity was overestimated using the slow motion footage. Even though the

Page 23/37
Theoretical and Experimental Investigation into the flight of an X-Zylo Nils Wagner

smaller launch velocity captured by the GoPro8 was accounted for in the calculation of
the initial parameters, due to a really large uncertainty the influence of that measurement
made a negligible influence in the weighted average. This shows that the uncertainty of the
launch velocity for the other cameras was likely underestimated.
Another detail that can be observed in the plot is that the x-velocity decreases more
rapidly shortly before hitting the ground as the sideways drift of the X-Zylo accelerates.
This can also be seen in the theoretical prediction where this behavior is starting to form
just as the ring hits the ground, therefore a bit later as in the observed trajectory. The
parabolic trajectory shows a constant x-velocity as no drag forces are applied.

23e) The z-velocity shows an interesting behavior. At launch the vertical velocity is quite large,
but as expected by the theoretic prediction in section 2, the ring at first loses height rather
quickly (first drop). The asymptotic behavior for t → 0 is seen to fit the parabolic trajectory,
which was predicted. Only after about 0.3 s this initial drop is absorbed by the increasing
lift force due to the rising AoA. The second drop is visible later during flight starting from
about 2.8 s onward. There the ring begins to lose height faster than seen before, which can
be explained by the flow separation forming at higher AoA. Later in plot 23o this becomes
more obvious when looking at the AoA. The mentioned equilibrium phase of the flight in
between both drop phases can be seen quite well using this representation.
Comparing theory and experiment, both curves match remarkably well until the second
drop phase sets in. During the second drop the decrease in z-velocity is underestimated by
the simulation, showing that the impact of the flow separation occurring was underrated.

23f) The velocity magnitude follows a similar pattern as the x-velocity. However, due to the
increasing negative z-velocity at the end of flight the velocity magnitude plateaus with
a tendency to rise again. A similar behavior is seen in the observed trajectory. Due to
additional effects stemming from the x- and y-velocity, the observed velocity magnitude
drops again shortly before hitting the ground. This effect is not visible in the theoretical
predictions and is not fully understood.

23g) This plot shows the observed x-velocity obtained by the slow motion footage (see figure 19).
As the velocity range is small and the uncertainties are fairly big, no error-bars were put
on the experimental data. When looking at the S10’s data, the theoretical behavior was
approximately met, however a dip is seen when the X-Zylo is in the center of the camera’s
field of view. This effect most probably occurs as some additional camera effects are not
corrected or are underestimated. As most probably the velocities observed in the center
of the field of view are more accurate, this would also backup the claim that the launch
velocity could be overestimated in the slow motion footage.
The data obtained by the S9 is even more questionable as a far faster drop in velocity
than expected is observed. Here the error stems from the position of the camera, as it
was seen in the footage that the camera was slightly tilted sideways, therefore producing
a skewed image. As both cameras are very close to the trajectory, the camera distortion
produces a relatively big error.
Another problem can be seen when looking at the plot itself. The experimental data
shows a quantized manner as only discrete velocities are seen, which is non-physical. The
resolution of the camera produces this pattern, as a pixelated image can only entail discrete
length differences. Therefore, one has to track the X-Zylo almost pixel-perfect, which
induces even more error.
It is quite obvious that the used method—capturing the launch of the X-Zylo with
cameras—to calculate the initial values accurately is not sufficient. Even though correction
terms are applied, the camera induced errors still dominate the obtained results. The closer
the camera is to the trajectory, the stronger those effects become.

Page 24/37
Theoretical and Experimental Investigation into the flight of an X-Zylo Nils Wagner

23h) The z-velocity matches the theory better as camera errors like barrel distortion and the
sideways tilt of the S9 mostly influence the x-coordinate. While the X-Zylo flight spans
almost the whole field of view horizontally, it only covers a small distance vertically. The
vertical components are only inflicted with serious errors for high launch angles.
The mentioned asymptotic behavior for t → 0 is best visible using this representation.
While the parabolic trajectory shows a linear decrease in z-velocity, the X-Zylo’s increasing
AoA dampens this decrease until a force balance is formed. This is the transition between
the first drop into the equilibrium phase.

23i) The dip in x-velocity observed by the S10 as well as the fast drop for the S9 are still visible
in the velocity magnitude plot. As the x-component of the velocity dominates during early
flight, the plot does not significantly alter from the observed early-time behavior of the
x-velocity itself seen in figure 23g.

23j) One can also look at the xz-projection of the velocity vector. The angle between the x-axis
and this projection is defined by α~v = arctan(vz /vx ) and will be denoted as velocity angle.
Using this quantity, the different flight phases are nicely visible, especially the transition
from the equilibrium phase towards the second drop. The first drop is less pronounced as
in plot 23e showing the z-velocity, however still visible. In this representation it can also be
spotted that the observed data only really diverges after flow separation occurs and the
second drop is initiated. The difference between theory and experiment in the plots 23d+f
across the whole flight duration can be traced back to the velocity magnitude, which was
likely overestimated from the slow motion footage. Using this representation the velocity
magnitude has no influence, only the ratio of x- to z-velocity is important. Therefore this
representation is most likely more accurate.

23k) When looking at the velocity angle in the slow motion footage, the first drop is observed
really well with both cameras. The theory and the experiment coincide almost perfectly
and again the asymptotic behavior for t → 0 matches free fall.

23l) The y-velocity also follows the expected contour, however there the first qualitative difference
between theory and experiment is seen. Instead of the y-velocity increasing linearly during
the second drop phase, the acceleration decreases and tends to zero, so that the velocity
plateaus. It is unknown if the velocity remains stable or reverses a second time, which
would then need a new explanation. With the given data no conclusion can be found for
this issue as especially the sideways drift is hard to model. The Transition-SST model is
seen to fit better to the real drift behavior than the k-ω-SST turbulence model.

23m) The angular frequency was only measured at launch and not during the rest of the flight.
Therefore, it is unknown whether the simplified model yields accurate results. Mind that
even though the behavior seems linear, this is not the case, especially for a longer flight
duration.

23n) The experimental values for the acceleration magnitude are not shown since they scatter
significantly as even small imperfections in the tracking amplify when deriving the quantities.
In the theoretical prediction the first drop is perfectly seen. As the X-Zylo does not generate
lift at start, the acceleration magnitude is 9.81 m/s (plus additional drag force). In the first
0.2 s this magnitude drops rapidly as the AoA of the ring increases. The behavior late in
flight is more complex and the different components of the acceleration would need to be
reviewed independently for a deeper analysis. Those are however not included here.

23o) At last the theoretical AoA of the ring is shown. Here also the different phases are illustrated
quite well. The ring starts off with an AoA of 0°; this condition was preimposed and an
assumption made for the launch with the launch device. In section 6.4 this assumption is

Page 25/37
Theoretical and Experimental Investigation into the flight of an X-Zylo Nils Wagner

discussed further. It is seen that as expected the AoA increases rapidly shortly after launch,
then increases only slightly during the equilibrium phase. After that the second drop sets
in, apparent from the rapid rise in AoA. This rise sets in after an AoA of approximately
10° is exceeded due to the flow separation forming, which was seen in figure 23. It is
perfectly visible that the effects seen in the CFD data necessarily also show in the simulated
trajectory.

All in all the behavior of the X-Zylo was qualitatively predicted very well. The different
phases mentioned in section 2 are clearly distinguishable in both the theory and the experiment.
Moreover, the duration of the different phases could also be predicted well. This in turn shows
that the CFD simulation was able to predict the flow separation qualitatively as already seen in
the validation cases. However, it is seen that even though the qualitative flight was predicted
correctly, especially the velocity of the ring at launch seems to be overestimated from the slow
motion footage. This can be seen when looking at the trajectory in the xz-plane (23a) and the
velocity magnitude plot for the whole trajectory (23f), which shows an offset for the whole flight
duration. It is theorized that amplified camera errors due to the proximity of the slow motion
cameras to the trajectory interfere with an accurate measurement of the initial parameters. Hints
for such behavior are seen in figure 23g with the observed dip in the center of the field of view.
An important detail is the asymptotic behavior seen for t → 0. Both the simulation and the
observed data asymptotically match the free fall simulation at launch, confirming a first drop. Not
only does this confirm the qualitative theory (see section 2.1), but also becomes very important
when studying a human induced launch. Resolving the additional effects one has to account
for in a human induced launch (see section 6.4), it will become clear that without the launch
contraption, a useful analysis of the flight would likely be impossible as exactly this asymptotic
behavior is violated.

6.2 Influence of the Launch Angle

As the launch angle proves to be an essential parameter in the flight of an X-Zylo and the launch
mechanism is build especially to control that factor, several launches with different initial launch
angles are compared. The initial launch values taken from the slow motion footage can be read
from table 3. As seen the launch angles varied from 8° to 15.5° with almost equal step size. The

launch angle initial velocity init. rot. freq. launch time

αlaunch [◦ ] vlaunch [m/s] ω(t0 )/(2π) [Hz] distance [m] aloft [s]
flat (L1) 8.0 ± 0.5 15.9 ± 0.3 47 ± 3 41.3 ± 0.3 4.02 ± 0.07
medium (L2) 10.3 ± 0.4 15.4 ± 0.3 42 ± 3 40.0 ± 0.3 4.23 ± 0.07
steep (L3) 13.0 ± 0.4 16.1 ± 0.2 60 ± 3 38.3 ± 0.2 4.47 ± 0.07
steepest (L4) 15.5 ± 0.5 15.4 ± 0.3 25 ± 2 32.1 ± 0.3 4.00 ± 0.07

Table 3: Initial launch values with uncertainty for the compared launches taken from the slow
motion footage. The launch distances and the flight time are included for comparison.

initial velocity however shows a variance depending on the launch. Also the rotational frequency
is very different for the launches which was already seen in section 4.2. The predicted trajectories
as well as the experimental data can be seen in figure 24 and 25. At first, it is seen that the
predicted trajectory shows a farther flight distance than observed in all cases. As seen in section
6.1, several effects play a major role for the underestimated flight distance. In particularly an
underrated flow separation for high AoA, a miscalculation of the launch velocity and a difference
in drift behavior are the most prone errors. The steepest launch L4 represents the underrated
flow separation really well, as the trajectory only deviates from the theory shortly before the
second drop sets in. The sideways drift behavior observed in the medium flat launch L2 seen

Page 26/37
Theoretical and Experimental Investigation into the flight of an X-Zylo Nils Wagner

in figure 25 shows, that the drift has non-negligible influence on the drop. As the ring drifts
less significantly into the positive y-direction compared to the simulated trajectory, the observed
drop is less rapid than simulated. Modeling the sideways drift more accurately could in fact also
improve the overall performance of the prediction as all flight directions are strongly coupled.

simulation (Transition SST) simulation (k − ω − SST) parabola comparison observed trajectory

4
αlaunch 8.0◦
2

αlaunch 10.3◦
z coordinate (height) [m]

0
6
αlaunch 13.0◦
4

6 αlaunch 15.5◦

0
0 10 20 30 40 50
x coordinate (distance) [m]

Figure 24: Comparison of the trajectories for different launch angles using equal axis for a better
view on the flight. Due to this scale the error-bars in z direction are scaled by a factor of 3 for
better visibility.

For the sideways drift of the trajectory (see figure 25), the observed values are still in the
confidence interval of the prediction made by the Transition-SST turbulence model, however only
due to the huge uncertainties given by the model. As additional information it is shown where the
y-acceleration vanishes and the torque on the ring flips (vertical lines). This happens when the
COP passes through the COM of the ring. For the observed trajectory an additional confidence
interval is shown as the torque flip is not sharply visible in the data. One can deduce that the
torque flip is mostly theorized too early during flight. The ring generally drifts stronger into
the negative y-direction before turning, showing that either the COM or COP data is erroneous.
Even minimal shifts in either parameter results in visible differences of the sideways drift (see
section 6.4).
Figure 26 represents the experimental observation of the velocity angle for all four launches.
The theoretical prediction again shows good agreement, however the deviation seen is larger than
formerly observed in figure 23e. This is especially the case for the equilibrium phase of the low
launch angle cases (L1, L2). Here the first drop lasts slightly longer than expected, therefore
creating a faster decline in the velocity angle than predicted. A good explanation was not found
as it would be expected that for lower AoA the prediction would be better as the CFD simulations
should yield more accurate results. In the case L4 the faster drop during late flight can be seen
again in the steeper slope of the velocity angle during the second drop phase.
Concerning the time duration of the different phases one clearly observes that as expected
the equilibrium phase gets shorter for higher launch angles. Here it comes into play that the
X-Zylo generates less lift in the z-direction for a steeper climb, therefore the AoA to support the

Page 27/37
Theoretical and Experimental Investigation into the flight of an X-Zylo Nils Wagner

simulation (Transition SST) simulation (k − ω − SST) torque flip observed trajectory

1.5 2.0
y coordinate (sideways drift) [m]

y coordinate (sideways drift) [m]

αlaunch 8.0◦ αlaunch 10.3◦
1.0 1.5

1.0
0.5

0.5
0.0

0.0
−0.5
−0.5
−1.0

COM > COP COM < COP −1.0 COM > COP COM < COP
−1.5
0 10 20 30 40 50 0 10 20 30 40
x coordinate (distance) [m] x coordinate (distance) [m]
1.5 3
y coordinate (sideways drift) [m]

y coordinate (sideways drift) [m]

αlaunch 13.0◦ αlaunch 15.5◦

1.0 2

0.5
1

0.0
0

−0.5
−1
COM > COP COM < COP COM > COP COM < COP
−1.0
0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 35
x coordinate (distance) [m] x coordinate (distance) [m]

Figure 25: Sideways drift behavior of the X-Zylo in flight for different launch angles. It is marked
where the COP moves past the COM on the chord and therefore the drift behavior changes. It is
shortly noted with COM > COP when the COM is in front of the COP, and vice versa. Note
that it is implicitly assumed that the COP always lies on the symmetry axis of the X-Zylo. This
assumption was used for all computations and is thought to sufficiently depict the real situation.

weight has to be higher. This increases the equilibrium AoA and therefore the drag on the ring
during the whole equilibrium phase. This can be seen as the velocity angle in the equilibrium
phase is higher for higher launch angles, which directly translates to the AoA as the ring axis
vector is virtually constant during the first two flight phases and only changes in the third phase.
The transition phase leading to the second drop is seen to be longer for small launch angles,
with the second drop phase being longer for higher launch angles as the ring gained more height
during flight. The equilibrium phase independently of the launch angle shows an almost uniform
duration. A notable detail in figure 26 is that in all launches, the asymptotic behavior of the
velocity angle for t → 0 again matches that of a parabolic trajectory. The deviation seen for the
first frames of the high launch angle cases is also small enough to still be in the given uncertainty
bounds. Therefore, the preimposed condition of an 0° AoA is approximately met.
Figure 27 shows a parameter analysis of the flight distance of an X-Zylo with set initial
parameters and variable launch angle. The flight distance with respect to the launch angle is
plotted for launch angles in the range [0°, 90°]. For comparison the curve for a parabolic trajectory
is shown, as well as the error margins for the flight distances. It is visible that instead of the
flight distance rising with higher launch angle, a small launch angle is favorable.

Page 28/37
Theoretical and Experimental Investigation into the flight of an X-Zylo Nils Wagner

simulation (Transition SST) simulation (k − ω − SST) parabola comparison observed values

10
αlaunch 8.0◦ 10 equilibrium phase αlaunch 10.3◦
equilibrium phase
5
velocity angle α~v [◦]

velocity angle α~v [◦]

0
0

−10
−5
2nd drop
1st drop 2nd drop 1st drop
8 10
−10 −20
6 8
−15
4
−30 6

−20
0.0 0.1 0.2 0.3 0.0 0.1 0.2 0.3
−40
0 1 2 3 4 5 0 1 2 3 4 5
time [s] time [s]

equilibrium phase αlaunch 13.0◦ equilibrium phase αlaunch 15.5◦

10 10

0 0
velocity angle α~v [◦]

velocity angle α~v [◦]

−10
−10
−20
−20 2nd drop
14 2nd drop
1st drop 16 1st drop
−30
−30 12
14
−40
10
−40 12
−50
8
10
−50 −60
0.0 0.1 0.2 0.3 0.0 0.1 0.2 0.3

0 1 2 3 4 5 0 1 2 3 4 5
time [s] time [s]

Figure 26: Comparison of the velocity angle of the ring during flight for different launch angles.
The first 0.35 s after launch are magnified and the different flight phases marked.

The initial values used for the parameter

simulation (Transition SST)
parabola comparison
simulation (k − ω − SST)
maxima
analysis were set to be
vlaunch = (16.0 ± 0.3) m/s ,
50
50
ω(t0 )/(2π) = (50 ± 3) rad/s ,
flight distance [m]

40
40
h(t0 ) = (1.40 ± 0.01) m .
6.5◦ ± 0.2◦

30
The uncertainty for the launch angle was
30
2.5 5.0 7.5 10.0

set to ± 0.4°. All other parameters were

20 set as if the colored X-Zylo was used. As
seen in figure 24, the flight distance de-
10
43.3 ± 0.3
creases for larger launch angles, however
◦ ◦

the maximum is predicted to be at a still

0
0 10 20 30 40 50 60 70 80 90 smaller launch angle of 6.5° ± 0.2°. Unfor-
launch angle αlaunch [°]
tunately no launch angles smaller than 8°
Figure 27: Parameter analysis of the flight distance were experimentally recorded during the
in respect to the launch angle for both turbulence final test run, therefore no comparison can
models and a traditional parabolic trajectory. be made for those cases. The parabolic
trajectory shows a deviation from the ex-
pected optimal launch angle of 45° due to the height offset at launch. It can be seen that the
X-Zylo operates well only in a small launch angle interval of about [3°,15°]. It is also seen that
the flight distance is much more sensitive to launch angles lower than the optimal value than to

Page 29/37
Theoretical and Experimental Investigation into the flight of an X-Zylo Nils Wagner

higher angles. This is due to two effects. On one hand a higher launch angle yields a smaller
equilibrium AoA, decreasing drag over an elongated flight period. However, for small launch
angles the ring cannot cover a long flight distance as it looses height during the first drop. This
leads to a trajectory that curves to the ground early in flight and therefore cannot fully utilize
the equilibrium phase. The optimal trade-off strongly varies depending on the ring mass, the
launch height and the initial velocity magnitude. As an in depth parameter analysis would go
beyond the scope of this paper, this is not covered in more detail.

6.3 Sideways Drift

The sideways drift is heavily dependent on the location of the COM. This is shown in figure 28,
where two different X-Zylo launches are shown. The initial conditions for both launches only differ
slightly (∆vlaunch = 2.8%, ∆αlaunch = 1.4%, ∆ω(t0 ) = 13.5%), however one X-Zylo is colored
uniformly while the second one is unaltered. As stated in section 5.1, coloring the X-Zylo shifted the
COM by 6% downstream, which has
simulation (Transition SST) simulation (k − ω − SST) a big influence on the observed flight
torque flip observed trajectory pattern. The unaltered ring turns sig-
2.5 nificantly earlier, which is an effect
y coordinate (sideways drift) [m]

4 colored comparison
2.0
of the COM being further upstream.
As predicted by the CFD data shown
2
1.5
0
in figure 10, the COP moves up-
1.0
−2
0 10 20 30 40 stream with increasing AoA. The po-
sitions in flight where a change in y-
6 0.5
uncolored
4 0.0
acceleration is seen are marked with
2 −0.5 the respective uncertainty. These are
0
−1.0 the positions where the COP moves
past the COM and the torque on the
0 10 20 30 40 0 10 20 30 40
x coordinate (distance) [m]
ring flips. An earlier torque flip for
Figure 28: Sideways drift behavior for the flight of a
the uncolored X-Zylo is seen as would
colored X-Zylo compared with a launch of an unaltered
be expected. As can be seen in figure
X-Zylo.
15, the flight is not very sensitive for
the rotational velocity of the ring, only the drift amplitude shows an almost linear dependence on
the initial rotational frequency. It is seen that the location of the COM is a much more sensitive
parameter.

6.4 Human Induced Launch

A core assumption made in all theoretical calculations is that the initial AoA of the ring is 0°.
However, it is seen in the work by Hirata, et. al. [4] that this assumption does not hold for a
launch by a human. To evaluate how the trajectory and the AoA behave for different initial
parameters, figure 29 shows trajectories for an initial launch angle of 5° and an initial velocity
magnitude of 16 m/s. The initial AoA is varied from −6° to 15° in discrete steps and the resulting
xz-projection of the theoretical trajectory shown. Additionally, the AoA is shown as it covers the
early flight in more detail.
It is seen that the trajectory of the ring bends upwards early in flight when the initial AoA is
very high. This can be explained using the model developed in section 2.1. As the AoA at launch
is higher than the equilibrium AoA, this produces a larger lift force than the opposite gravitational
pull and therefore the ring is accelerated upward. This then results in a smaller AoA and returns
the ring to the equilibrium AoA over time. The first flight phase is therefore seen to be a first rise
of the ring instead of a drop. It is visible that the ring rapidly moves towards an equilibrium AoA
no matter the initial AoA. From there on the flight is normal again, only the first flight phase
is altered. However, it can be seen that even a change in initial AoA of 3° gives tremendously

Page 30/37
Theoretical and Experimental Investigation into the flight of an X-Zylo Nils Wagner

different results in the trajectory. Moreover, one can observe that the equilibrium AoA is different
for the different initial AoA values. This effect is due to the different velocity magnitudes the
ring has when encountering the equilibrium. As the drag is approximately rising quadratically for
higher AoA, the trajectory with an initial AoA of 15° shows a higher equilibrium AoA as the
ring lost more speed early on in the first rise. Additionally, with the trajectory bending upward,
again the lift does contribute less towards maintaining a straight flight as the z-component is
diminished. As already stated in section 6.2, it is again seen that the first flight phase before an
equilibrium is achieved lasts approximately 0.25 s no matter the initial AoA or the launch angle.
Similar to a pendulum a stronger deviation from the equilibrium position results in a greater
restoring force.

initial AoA at launch α(t0 ) [°]

6
2.2 15 15
z coordinate (height) [m]

*
2.0 1st rise

angle of attack α [°]

5 12
1.8 10
9
4
1.6
6
3 1.4 5
0 1 2 3 4 3
2 equilibrium phase
* 0 0

1 1st drop -3
−5
0 -6
0 5 10 15 20 25 30 35 40 45 50 0.00 0.25 0.50 0.75 1.00 1.25
x coordinate (distance) [m] time [s]

Figure 29: Comparison of trajectories with different initial AoA. A 0° initial AoA is the core
assumption of all former calculations. However, when the ring is launched by a human, this
assumption must not hold, causing different phenomena to occur.

A real human-induced launch is visible in figure 30, carried out by Hirata, et. al. [4]. It is
visible that the flight behavior is remarkably different from the observed behavior with the launch
device. The X-Zylo rises after launch
3
before encountering the equilibrium
phase, just as seen in figure 29. When
z-coordinate (height) [m]

2 trying to reconstruct the found tra-

jectory one finds good agreement for
1 a negative launch angle of −10◦ as
well as for a high initial AoA of
0
about 21◦ . Those values—even if
only approximate—show the signifi-
observed trajectory Hirata, et al [5] cance of a launch device, with which
−1
simulation (k − ω − SST) the initial AoA can be controlled. As
simulation (Transition SST) seen in figure 29, an initial AoA of
−2
0 5 10 15 20 25 30 35 40 45 50 more than 10° would lead to an ex-
x-coordinate (distance) [m] tremely fast rise of the X-Zylo. Be-
cause of that, experimentally eval-
Figure 30: Launch by a human recorded by Hirata, et.
uating the initial parameters would
al. [4]. The launch can be modeled with a high initial
become impossible, as the velocity
AoA of approximately 21° and a negative launch angle
vector used to find the launch angle
of about −10°.
changes drastically in the first 0.1 s
of flight. Additionally, the initial AoA of the ring itself is hard to determine. As significant initial
parameters cannot be determined accurately, a quantitative comparison between theory and
experiment is unlikely to be of good quality. One could only use a maximum likelihood analysis
to determine the most suitable initial parameters, however without knowing the experimental
parameters, a comparison is still not possible.
It has to be emphasized that as in all launches conducted using the dedicated launch device,
the asymptotic behavior for t → 0 was approximately that of a parabolic trajectory. However,

Page 31/37
Theoretical and Experimental Investigation into the flight of an X-Zylo Nils Wagner

the implications of a non-zero initial AoA as seen in figure 29 would be strongly visible. This
shows that the condition of an initial 0° AoA is suitably met with the launch apparatus, as even
small deviations would result in a different asymptotic behavior.

6.5 Open Issues

Several problems still prevail, which have to be considered for further analysis of the flight. A
non-comprehensive list with some open issues is described in the following:

i) CFD Data: The CFD Data is seen to be the weak point of the theoretical model. Even
though the model was validated and even rotational effects were taken into consideration,
still a deviation can be seen especially for the high AoA flight. Here more sophisticated
methods as for example LES simulations would be needed for better coverage of the effects.
Also in contrary to what was thought before, the COP is a very sensible parameter and has
to be known with high precision. Therefore, the medium M2 mesh with its 3.5% (k-ω-SST)
or 6.9% (Transition SST) error from the finest mesh is not accurate enough. A better
resolution here would be needed for this parameter as well as an additional validation with
experimental COP data. Moreover, the assumption that CD and CL do not depend on
the flow velocity has to be investigated. As was seen in figure 23f, the velocity magnitude
during flight ranges from 6 m/s to 16 m/s. In contrast, the CFD analysis was performed for
a flow velocity of 17.15 m/s, which could yield more slight deviations that add up over time.
Also, the influence of the model’s rotation was only calculated for an AoA of 5°, a different
behavior could be present for other AoA.

ii) Launch Device: To alter the launch device for higher reproducibility would be far
fetched, only the belted motor drive mechanism could be improved easily to retain a more
stable rotational frequency. However, as the initial values are calculated for every launch
individually, reproducibility is not seen as a key factor needed for better experiments. It is
more important to satisfy the initial 0° AoA condition needed to make good predictions. In
former iterations of the launch mechanism, this condition was not met and strong effects as
seen in figure 29 were visible. Those issues were resolved, however it is not known what the
deviation from the condition at this final point really is. As it is not feasible to observe an
initial AoA of less than 1°, the only possibility would be to further improve the device to
ensure the condition is met accordingly all the time.

iii) Tracking the Trajectory: As five different cameras were used with non-identical lenses
and intern software it is impractical to correct the trajectory reliable. The used corrections
as stated in 5.2 only concern geometrical corrections, camera induced errors were neglected.
At best one would use two synchronized high resolution cameras that observe the trajectory.
With additional markers on the X-Zylo one would then be able to use more advanced
tracking software with computer vision algorithms to probe the flight accurately. This
would then also ensure reproducibility as a set algorithm performs the tracking and correction
procedure rather than a human tracking frame by frame manually. However, this would
need a dedicated camera setup currently not available to the author. Future improvements
in LiDAR (Light Detection And Ranging) technology could also be a suitable tracking
alternative, bypassing some imaging errors one encounters with cameras.

iv) Initial flight parameters: The initial flight parameters as well as their uncertainty are
key factors for the theoretical prediction of the flight behavior and are seen as the biggest
error source involved. Using close-up cameras, only the rotational frequency is reliably
accessible; the initial launch velocity and the launch angle can only be estimated roughly.
Several factors as for example the close proximity of the cameras to the trajectory and
tracking the object manually hinder this process. If one tracks the X-Zylo frame by frame

Page 32/37
Theoretical and Experimental Investigation into the flight of an X-Zylo Nils Wagner

in the slow motion footage, the time resolution is very high. However, as the distance
traveled is only some pixels wide, the velocity deviates much and shows a discrete spectrum.
If however only every tenth frame is probed, the time resolution is to small to observe
the first drop in its full extent since the initial quantities change rapidly after launch. For
example in the experiments the initial launch angle changed about 1.5° in the first 0.1 s
aloft (see figure 26). Additionally, camera distortions and other geometric considerations
mentioned in section 5.2 due to the proximity of the camera further increase the uncertainty
on the gained values. Therefore, it is necessary to either use a dedicated high-speed camera
setup for the launch or to use alternate means of probing the initial values. One could
think of several light barriers in close distance to the launch origin to probe the velocity
reliably. The launch angle could theoretically be measured using the central guiding rod of
the launch mechanism. For now this was not done as the stability of the device does not
allow for accurate measurements using the apparatus itself.
v) Incomplete Theoretical Model: The theoretical model is known to be incomplete as
minor effects were not accounted for, for example a small Magnus force when drifting
sideways, therefore having a small velocity component perpendicular to the ring’s axis. It
was also preimposed that the COP is always located on the ring’s symmetry axis, which
does not have to be true. As those additional effects should have insignificant influence on
the flight, they were neglected. However, it is not certain if for different flight scenarios
those effects become important, e.g. for crosswinds outside. There could also be additional
forces and torques at play that were not considered.

7 Conclusion
It was found that the conceptual explanation of the flight as well as the theoretical simulation
can predict and explain the observed flight behavior in great detail. Only minor details like the
plateauing y-velocity during late flight are not yet fully understood. However, it is not unlikely
that those remaining anomalies are an effect of the imperfect camera setup used. The flight of an
X-Zylo was seen to be very sensitive to several initial parameters, especially the initial launch
angle, the initial AoA, and the location of the COM. Particularly the initial AoA, which can be
kept close to 0° with a launch device, makes a quantitative analysis of the whole flight almost
impossible when launching the X-Zylo by hand. Therefore, even without good reproducibility,
the launch mechanism still is a key part of the experiment as it allows to neglect the impact of a
critical parameter.
From a quantitative point of view the flight was found to be modeled sufficiently precise,
however several areas have to be improved upon. At first the second drop cannot be captured
in its full extent as the flow separation strongly impacts the second half of the flight. It is seen
that the CFD data can predict the moment of separation correctly, however the impact of the
separation is underestimated. Therefore, most simulations show a longer flight distance and a
less significant drop at the end of flight compared to the observed data. Another observation is
that the sideways drift is stronger than predicted, therefore the COM and COP locations have to
be calculated more accurately. Another weak point is the used camera setup. A more dedicated
setup with additional calibration points and an automated tracking procedure would be needed
to advance the experimental measurements. Especially during early flight it is hard to capture
the velocity and the initial launch angle with great detail, as both values change rapidly in the
first 0.2 s of flight.
All in all it is expected that advancements in the CFD data and the camera setup would
entail a remarkable accordance between theory and experiment. While a qualitative parameter
analysis can indeed be made with the given setup (see figure 27), a detailed quantitative analysis
can not yet be achieved. Therefore, the goal of this work was only partially achieved with further
work being necessary.

Page 33/37
Theoretical and Experimental Investigation into the flight of an X-Zylo Nils Wagner

Acknowledgments
I want to thank the whole SciComp research group at the TU Kaiserslautern, especially Ole
Burghardt and Tim Albring, for their correspondence and indispensable help to start this project.
Additionally, I would like to thank CADFEM as well as Dr. Thomas Frank and the LRZ
for providing the license for the ANSYS software and access to High Performance Computing
resources. Furthermore, advice given by Prof. Dr. Nicolas Gauger and Prof. Dr. Christian
Breitsamter regarding the publication of this work was greatly appreciated. I wish to extend my
special thanks to Hermann Steffen and Christopher Reinbold for their insightful comments on
earlier versions of this manuscript.

8 References
[1] William Mark Corporation (online). Accessed on March 20, 2020. url: https://www.
wmctoys.com/products/x-zylo.
[2] William Crowther and Jonny Potts. “Simulation of a spinstabilised sports disc”. In: Sports
Engineering 10 (Jan. 2007), pp. 3–21. doi: 10.1007/BF02844198.
[3] Mont Hubbard and S. Hummel. Simulation of Frisbee Flight. Jan. 2000. url: https :
//www.researchgate.net/publication/253842372_Simulation_of_Frisbee_Flight.
[4] Katsuya Hirata et al. “Field observation and numerical analysis of a rotating pipe in flight”.
In: Journal of Fluid Science and Technology 13.3 (2018). doi: 10.1299/jfst.2018jfst0021.
[5] David London Tarr. What makes the Amazing X-Zylo fly. 1.0. Columbia-Capstone, 2016.
isbn: 978-0-9821148-3-4.
[6] Peter Kämpf. Accessed on March 20, 2020. 2016. url: https://aviation.stackexchange.
com/questions/24782/how-does-a-ring-paper-airplane-fly-for-a-long-distance.
[7] H. Schlichting and E.A. Truckenbrodt. Aerodynamik des Flugzeuges: Erster Band: Grundla-
gen aus der Strömungstechnik Aerodynamik des Tragflügels. 3rd ed. Klassiker der Technik
Teil 1. ISBN: 978-3-540-67374-3. Springer Berlin Heidelberg, 2001. isbn: 9783540673743.
url: https://www.springer.com/de/book/9783540673743.
[8] The MathWorks, Inc. MatLab. Version R2018b. Sept. 13, 2018. url: https://mathworks.
com/products/matlab.html.
[9] Herbert S. Ribner. “The Ring Airfoil in Nonaxial Flow”. In: Journal of the Aeronautical
Sciences 14.9 (1947), pp. 529–530. doi: 10.2514/8.1437.
[10] Svetopolk Pivko. “Zur Abschätzung der aerodynamischen Eigenschaften dünner kreiszylin-
drischer, schrägangeströmter Ringflügel”. In: ZAMM - Journal of Applied Mathematics
and Mechanics / Zeitschrift für Angewandte Mathematik und Mechanik 36.7-8 (1956),
pp. 306–307. doi: 10.1002/zamm.19560360746.
[11] J. Weissinger. Zur Aerodynamik des Ringflügels. Die Druckverteilung dünner, fast drehsym-
metrischer Flügel in Unterschallströmung. Vol. 198. Forschungsberichte des Wirtschafts-
und Verkehrsministeriums Nordrhein-Westfalen. ISBN: 978-3-663-04040-8. VS Verlag für
Sozialwissenschaften, 1955. isbn: 9783663040408. doi: 10.1007/978-3-663-05486-3.
[12] Open Cascade SAS. SALOME. Version 8.3.0. Sept. 6, 2017. url: https://www.salome-
platform.org/.
[13] ANSYS, Inc. Fluent. Version 2019.R3. Sept. 10, 2019. url: https://www.ansys.com/
products/fluids/ansys-fluent.

Page 34/37
Theoretical and Experimental Investigation into the flight of an X-Zylo Nils Wagner

[14] Herman S. Fletcher. “Experimental Investigation of lift, drag and pitching moment of five
annular airfoils”. In: NACA TN 4117 (Oct. 1957). url: https://ntrs.nasa.gov/archive/
nasa/casi.ntrs.nasa.gov/19930084906.pdf.
[15] Lance W. Traub. “Experimental Investigation of Annular Wing Aerodynamics”. In: Journal
of Aircraft 46.3 (2009), pp. 988–996. doi: 10.2514/1.39822.
[16] Lance W. Traub. “Experimental Study of a Morphing Annular Wing”. In: Journal of Aircraft
56 (Sept. 2019), pp. 1–7. doi: 10.2514/1.C035600.
[17] Edmund V. Latoine. “Wind Tunnel Tests of Wings and Rings at Low Reynolds Numbers”. In:
Fixed and Flapping Wing Aerodynamics for Micro Air Vehicle Applications. 2001, pp. 83–90.
doi: 10.2514/5.9781600866654.0083.0090.
[18] Michael J. Werle. “Aerodynamic Loads and Moments on Axisymmetric Ring-Wing Ducts”.
In: AIAA Journal 52.10 (2014), pp. 2359–2364. doi: 10.2514/1.J053116.
[19] Douglas Brown and Wolfgang Christian. Tracker. Version 5.0.6. Aug. 15, 2018. url: https:
//physlets.org/tracker/.

9 Appendix
9.1 Error Approximation for the Simulation
An important step in the trajectory simulation is calculating an uncertainty corridor in which
the X-Zylo is predicted, using the given uncertainties measured in the experiment. Therefore, a
rough error approximation is necessary, as all the initial parameters for the launched X-Zylo as
well as the drag and lift data are afflicted with error.
The trajectory is calculated using a function f , which inputs are all the initial parameters
{a, b, c, ...} and which output is the trajectory (and auxiliary information):
P~com (x(t), y(t), z(t)) = f (a, b, c, ...) .
Let us consider an initial uncertainty ∆a for the parameter a. The used approach is to call f
three times, once with the parameter a and then twice with parameters a ± ∆a. This yields three
different trajectories:
h i
P~com = f (a, b, c, ...) ,
h i 0
~
Pcom = f (a ± ∆a, b, c, ...) .
±∆a

This is done for every parameter with an inscribed uncertainty  while the time step
 ∆t is
~ ~ ~
kept constant so that one ends up with many trajectories Pcom 0 , Pcom ±∆a , Pcom ±∆b , . . . .
 

Therefore, it is now possible to calculate a vectorial difference between the baseline trajectory
P~com 0 and the ones induced with error for each time step tn
 

 
h i h i h i ∆x(tn )
∆P~com (tn ) = P~com (tn ) − P~com (tn ) = ∆y(tn ) ,
±∆a ±∆a 0
∆z(tn ) ±∆a
which is also calculated for every other trajectory, labeled with corresponding subscripts. Now
the approximate maximum uncertainty at time tn —assuming independent uncertainties for the
initial parameters—for e.g. the x-coordinate is calculated to be:
r v X
h i   2   2 u   2
∆x(tn ) = ∆x(tn ) ±∆a + ∆x(tn ) ±∆b + ... = u t ∆x(t )
n ±∆i >0.
±tot
i∈{a,b,c,...}

Page 35/37
Theoretical and Experimental Investigation into the flight of an X-Zylo Nils Wagner

To not mix between trajectories over- or undershooting the baseline trajectory P~com 0 , it is
 

checked that all differences contributing to the errors ∆±tot are of the same sign. Consequently
x(tn ) ± ∆x(tn ) ±tot are upper and lower
 

bounds for the x-coordinate of the X-Zylo

at time tn . Figure 31 shows all six limiting
points that get calculated and in which bounds 
the X-Zylo is most likely to be found during
 
P~com (tn ) 0 ∆z(tn ) +tot

flight. One could now lay a generalized ellip-  
soid through those points, however for the sake ∆y(tn ) +tot
of simplicity the bounding box around those
points is used as the approximation. In the
2D case of looking at the trajectory in the xz-  
∆y(tn ) −tot
plane, the upper and lower trajectory limits  
∆z(tn ) −tot
will be the points
h i h i h i
P~com (tn ) = P~com (tn ) + ∆P~com (tn ) ,    
upper 0 +tot ∆x(tn ) −tot ∆x(tn ) +tot
h i h i h i
P~com (tn ) = P~com (tn ) − ∆P~com (tn ) Figure 31: Visualization of the uncertainty
lower 0 −tot
approximation at time tn .
as they are the corners of the bounding box.
This approach can be generalized beyond the trajectory to other parameters that are outputted
by the simulation, however those cases are easier due to them being plotted against time rather
then other parameters also inflicted with error.

9.2 Error Approximation for the experimental Trajectory Observation

Not only the simulation is inflicted with error, also the usage of cameras to observe the flight’s
trajectory gives erroneous information. In section 5.2, it is touched upon how the observed
trajectory is corrected and which imaging errors are considered. However, as some camera induced
errors were left out of the calculation and the applied corrections still contain small uncertainties,
the influence of those effects has to be approximated. The calculation is only shown for the
X-Zylo’s position measurements, the velocity errors were calculated similarly. Several error sources
for the location of the X-Zylo are seen as independent, each afflicted with a certain error:

i) Uncertainty while tracking: This error is purely human-induced. As the X-Zylo is

only seen as a small black blob in the tracking software, it is hard to accurately track
the ring. Especially when the background changes color—e.g. when the ring flies in front
of a basketball net— the position is only vaguely visible. In addition, manual tracking
is not perfect and not the same spot is tracked every frame, which further increases the
uncertainty. Still this error is relatively small compared to other uncertainties. All in all a
1 cm uncertainty is added due to this tracking procedure for all coordinates.

ii) Barrel distortion: Even though both GoPro’s shoot in linear mode, a very small barrel
distortion is still visible which affects the calibration scale as well as the observed position
of the X-Zylo. The error on the scale is discussed further in point iii), only the error on the
tracked position is approximated here.
As the barrel distortion effects increase towards the edges of the camera’s field of view, the
added uncertainty depends on the difference between the position of the camera xcamera
and the tracked object xcorrect . A linear error term is then added

δxdistortion = |xcorrect − xcamera | · Rdistortion (18)

with the error rate Rdistortion . This rate has to be approximated for every camera used
as well as every coordinate due to the barrel distortion acting differently for different
coordinates. The approximated rates can be read from table 4.

Page 36/37
Theoretical and Experimental Investigation into the flight of an X-Zylo Nils Wagner

iii) Scale error: As already mentioned in point ii), the small barrel distortion adds an
uncertainty to the calibration scale used. From measuring each calibration segment it was
found that the uncertainty is under 1% for the GoPro8 and about 4% for the GoPro4.
However, it was also seen that the larger error of the GoPro4 does not stem only from the
scale error. Only about 1% again is from the scale error, the rest is due to error v). Using
this scale uncertainty an additional linear error term

δxscale = xcorrect · Rscale (19)

is added to the whole uncertainty. The scale error Rscale is dependent on the camera itself
and on the rotation of the camera.

iv) Error due to X-Zylo’s drift: As can be seen in equation (17) the x-correction involves—
additionally to the observed X-Zylo position—the different camera locations seen in figure
16. As those were measured beforehand, even a large measurement error of about 0.1 m
would add a negligible error, however the value ∆d is inflicted with greater error as it has
to be taken from the measurement of another camera (S7). As no dedicated equipment was
used and both cameras were not synchronized, an error of 10% for ∆d is used.

v) Camera rotation: It has a big influence if the camera is rotated slightly and is not set
perfectly perpendicular to the flight path. This also influences the scale, again due to the
intercept theorem. Therefore, if it is visible in the footage that the camera is turned slightly,
the scale error factor Rrotation has to be added to the value Rscale as seen in iii).

Summarized we get a total uncertainty on the position of the X-Zylo, e.g. the x-coordinate is
afflicted with an approximate error of

δxtotal = δxtracking + δxdistortion + δxdrift + δxscale + δxrotation (20)

|∆d|
= 1 cm + |xcorrect − xcamera | · Rdistortion + |xcorrect − xcamera | · dcamera · (0.1 + Rrotation ) + xcorrect · Rscale .

Note that all errors and free parameters are only estimated based on the footage to make an
educated approximation of the real uncertainty. A better consideration of the uncertainties is
needed.

Rdist,x Rdist,y Rdist,z Rscale,x Rscale,y Rscale,z Rrot,x Rrot,y Rrot,z

GoPro8 0.005 0.005 − 0.010 0.010 − 0.05 0.01 −
GoPro4 0.008 0.008 − 0.015 0.015 − 0.50 0.10 −
S7 − − 0.005 − − 0.015 − − 0.05

Table 4: Estimated values for all error rates. (Rdist = Rdistortion and Rrot = Rrotation )

Page 37/37