Background:
Despite its geometrical simplicity the FRISBEE (disc) gives rise to complex
flow patterns during flight. It is its inherent complexity that has made it a
popular toy and has
also allowed an entire sport to be built around it.
The desired properties of the disc during the play of Ultimate vary as it is a
very dynamic sport with the players constantly changing position, trying to
get ahead of their
mark. Consequently the players have developed a variety of different throws,
each of
which are suited to specific scenarios. Thus it is very difficult to say which
aerodynamic
properties the disc should have but the properties which have the most
influence on its
flight are:
• Coefficient of lift
• Coefficient of drag
• Sensitivity of these properties against varying angles of attack
and rotational.
The basic design of the FRISBEER has changed very little and indeed the
requirements
of the disc which may be permitted to be used in competitive Ultimate are
very strict
and do not allow for much variation . However, in recent years many new
designs
have hit the market, most notably the Aerobie Superdisc.
The solver calculates the flow through the region using a finite volume
technique which
integrates the governing equations of the flow in all the cells in the wind
tunnel. A process of discretization then gives values to the terms within the
integrated equations by finite-difference substitution. A number of different
solvers may be selected to calculate the flow field by using either a direct or
iterative method.
The post-processor is then used to display and analyze the results as well as
to determine their validity. The post-processor provides the user with a large
variety of tools with which to view the flow through the wind tunnel.
• Validate the data generated using CFD against actual wind tunnel tests and
in
flight measurements.
• Compare the different flow fields and identify the geometric features which
most
impact on the aerodynamic properties of the disc.
Plan of Development:
This report begins with a discussion on the literature that was available
regarding this
subject. The geometries of the discs are then discussed on how they were
modeled and
meshed in GAMBIT . This is followed by an explanation of the solver settings
which
were specified. The results are then shown, followed by the conclusions
which were
drawn and then by the recommendations which were made.
Lift and drag acting on a disc for a typical angle of attack. What sets the
flying disc apart from standard aerofoils is that it rotates during its flight,
indeed the rotation is absolutely critical for a stable flight.
The center of lift for a rotating disc does not coincide with its center of mass
and indeed
its position varies depending on the angle of attack. The result is that the lift
produces
a pitching moment about its center of mass.
Without the rotation this unbalanced pitching moment will dominate the
flight and
force the nose up.
The effect of the rotation causes the disc to effectively act as a gyroscope.
The unbalanced pitching moment coupled with the rotation then causes the
disc to bank or
roll to the left or right depending on the direction of the rotation. Thus the
flying disc
has an inherent instability built into its design, however, this has only served
to further
complicate the flight physics of the disc and made the sport of Ultimate that
much more
interesting, providing further options for the thrower.
• Pre-processor
• Solver
• Post-processor
The Pre-processor:
The pre-processor enables the user to create the geometry that he wishes to
model and
to create the mesh surrounding the model. The mesh compromises of a large
number of
cells and it is up to the user to carefully specify the mesh in order to
accurately capture
the flow phenomena and to minimize the time required for the solution to
converge. A
fine mesh is required in regions where large gradients are expected but is
unnecessary
in regions where the flow is stable and where a coarser mesh would be
adequate.
A wide variety of meshing schemes are available as well as differing cell
geometries,
each with its own inherent strengths and weaknesses. The numerical errors
which are
associated with CFD are closely linked to the shape and skewness of the cells
within the mesh. The meshing schemes can be divided into two categories,
namely structured and unstructured meshes.
Structured meshes allow the user to exactly specify the mesh and the cells
therein and
are preferable as they provide consistent cell geometries which can be
controlled and
thus numerical errors are minimized.
Unstructured meshes are best used when a structured mesh cannot be
applied and in
areas with low gradients. The unstructured meshing schemes adapt the
mesh in order
to fit the geometry of the model and the meshes of adjacent volumes. If not
carefully
implemented and controlled these schemes can result in highly skew cells
which can
exacerbate numerical errors.
Four different cell geometries are available, each with their own strengths
and weakness and it is up to the user to specify where they should be used.
Hexagonal cells are preferable especially in areas with high gradients
however they can only be used in structured meshes. Pyramidal cells are
also available and are used with unstructured meshes, they do however
result in additional numerical errors. The five sided pyramidal cell is used
when the mesh is transitioning from hexagonal cells towards the four sided
tetrahedral cells.
Due to the limitation of current computer speeds it is necessary for the user
to carefully
specify the mesh and the cells, fine enough in regions of high gradients to
accurately
capture the flow while being aware that unnecessarily fine meshes only
serve to increase the time required for a solution to converge.
The Solver:
The CFD software used for this thesis utilized the finite volume method as a
means of
solving the fluid flow. However all of the available methods follow three basic
steps:
The Post-Processor:
Turbulence Modeling:
The Spalart Allmaras model also utilizes the Boussinesq hypothesis and only
uses one transport equation, this lowers the computational time required for
a solution
to converge.
Extremely high gradients are experienced in the buffer layer where the flow
interchanges from laminar to turbulent flow. As such an extremely fine mesh
is required to fully capture the flow. However a coarser mesh may be applied
and a wall function can be used, which utilizes semi-empirical formulae
which are used to bridge the viscosity affected region between the wall and
the fully-turbulent region. As such the viscous sublayer and buffer layer are
not resolved and much computational expense is spared.
Also the Spalart Allmaras model was designed to be utilized for low Reynolds
number
applications, and consequently with models that properly resolve the viscous
affected
region. Therefore a near wall model should be used, although the wall
function approach
may still be utilized. It is recommended that with the application of the near
wall model
that the y+ should be approximately 1 but values lower than 5 are still
considered good.
The front end of the wind tunnel as well as the walls on each side where
defined as
Velocity Inlets, in the boundary condition settings in FLUENTR . The velocity
could
then be specified component wise and by altering the components,
effectively different
angles of attack could be investigated by using only one mesh. Care had to
be taken to
ensure that the wind tunnel was large enough so as the walls would not
adversely effect
the air flow.
From the initial simulations that were run, it became clear that the size of the
wind
tunnels could be reduced, so as to reduce the number of cells in the mesh
and thus
lower the computational time required for the solutions to converge. The
wind tunnel
geometries were kept constant once a suitable size was found, this was done
so as to
ensure consistency of the results.
The front and side faces of all the wind tunnels were specified as Velocity
Inlets, this
allowed the user to effectively alter the angle of attack by specifying the
velocity component wise.
The back faces were specified as pressure inlets as this would have the least
effect on the airflow as it exited the wind tunnel. A velocity outflow boundary
condition can effect
the air flow to force consistent exit vectors, however a pressure inlet
condition will not
interfere with the air exiting the wind tunnel.
The nature of the face mesh resulted in the growth of the cells from the inner
circle
towards the outer perimeter. It was required to have a fine mesh on the
outer perimeter
and as such, the density of the mesh grew towards the center of the disc. As
a result
an unnecessarily fine mesh was thus applied over the center of the disc,
however it was
found that this was unavoidable.
The greatest discrepancies in cell sizes occurred between the cells on the
surface of the
disc, which had a very large area when compared to their depth. Their area
was relatively larger than their depth and this was done deliberately in an
attempt to reduce the number of cells near the center of the disc. However
the depth of these cells was kept very small as was required in order to
accurately capture the boundary layer. As such GAMBIT could not generate a
volume mesh surrounding these volume, using any of the schemes available.
Hexagonal face meshes were applied to the disc, with a dense mesh near the
surface of the disc in order to capture the turbulent region. The mesh was
grown to form a shallow cylinder surrounding the disc, consisting entirely of
hexagonal cells.
The area close to the disc is of interest to the flow, and consequently
requires a dense and accurate mesh. However further away from the disc
the flow has very small gradients and as such, a tetrahedral mesh may be
applied while not affecting the flow and can save on computational time. As
such tetrahedral meshes were used to mesh the wind tunnel , in conjunction
with intermediate rectangular volumes, which aided the growth acceleration
of the tetrahedral cells.
Incompressible Flow
The velocities of the discs considered for this investigation were first at
10m/s. This velocity is well below 0.1 of Mach Number and thus the air flow
was assumed to be considered incompressible, this is considered acceptable
CFD
practice.
• Density 1.225kg/m3
• Cp(Specific Heat) 1006.43/kg-k
• Thermal Conductivity 0.0242w /m-k
• Viscosity 1.7894e-05kg/m-s
• Molecular Weight 28.966kg/kgmol
Operating Pressure
The operating pressure was specified as standard atmospheric pressure of
101.325kPa.
Turbulence Model
The Spalart-Allmaras turbulence model was used for this investigation, it
offers several
advantages over other turbulence models for this particular application. This
turbulence
model was designed to be used for low-Reynolds-number flows as such is the
case with
this investigation.
The flow over the discs for the higher velocity of 20m/s would be largely
turbulent, even
when the flow was 6m/s there would be turbulent regions within the flow
field. As such,
utilization of a laminar model would be inappropriate, and so the Spalart-
Allmaras turbulence model was specified, as it was initially designed to
handle exterior flows.
Convergence Criteria
Residuals are used to monitor the level of convergence within the flow for an
iterative
simulation. The total residuals are calculated as the difference between the
values of
those entering and leaving the volume.
The residuals used for this investigation were continuity, velocity in the x, y
and z directions
and μt which is the turbulent viscosity, which is introduced by the Spalart
Allmaras
turbulent model. Initially residual levels were set to 1e-3 and these were
satisfactory for
most simulations.
Setting the residuals to 1e-4 or lower would have led to prohibitively long
simulations
when computational resources were already severely limited.
Discretization Scheme
Several solution algorithms are provided by FLUENT in order to solve the
pressure velocity coupling in steady flows. For the initial simulations the
SIMPLE discretization
scheme was used. SIMPLE is a first order upwind scheme, however the
results that were produced did not match experimental data.
Scaled Residuals
Drag Convergence
Lift Convergence
Solution Converged
Profiles of Modified Turbulant Viscosity
Case 3 Angle of attack 20 deg, with the rotational speed
20 rad/s and air velocity of 10 m/s.
Scaled Residuals
Drag Convergence History
Values of Interest
The values that are of most interest for this investigation are:
• Co-efficient of Lift
• Co-efficient of Drag
The equations which define the co-efficient of lift and drag are shown below.
Case 1:
Case 2:
Cd = F/(0.5)(1000)(pi)(112²)(10²) = 64782.652/1970406912 = 3.29
Case 3:
Cd = F/(0.5)(1000)(pi)(112²)(10²) = 461.41867/1970406912 = 2.34
Conclusion:
In contrast to the major change in boundry layer and immediate flow field
surrounding a spinning disc, the overall aerodynamics were minimally
affected, if we exclude the gyroscopic stability benefit of rotation.
Although the numerical simulations were limited to a single shape and only
one variation in flight parameters, the overall CFD analysis was able to unveil
most of the “big picture” in disc aerodynamics. The influence of rotation on
boundry layer growth and flow separation was anything but expected, and
trying to obtain the same level of documentation in an actual wind Tunnel
test would be next to impossible. Although the number of graphs generated
from the computational data may have been overwhelming, well collectively
they mapped out the flow field and thus told us the “ complete story”.