Frisbee-circular plate aerodynamics

Discussion in 'Hydrodynamics and Aerodynamics' started by Mikko Brummer, Apr 18, 2012.

  1. Mikko Brummer
    Joined: May 2006
    Posts: 574
    Likes: 83, Points: 28, Legacy Rep: 258
    Location: Finland

    Mikko Brummer Senior Member

    I finally took the time analyse, on Leo's suggestion, the frisbee aerodynamics.
     

    Attached Files:

  2. Leo Lazauskas
    Joined: Jan 2002
    Posts: 2,696
    Likes: 155, Points: 63, Legacy Rep: 2229
    Location: Adelaide, South Australia

    Leo Lazauskas Senior Member

    The lip of the frisbee makes it a very tough problem.
    That's why I suggested you try a simple flat circular plate first.

    My contribution (with E.O Tuck) was to try to find cambered circular wings
    with zero pitching moment. The preprint of the paper which appeared in J. Ship Research can be found at:
    http://www.cyberiad.net/library/pdf/tl150104.pdf

    Good luck!
    Leo.
     
  3. Mikko Brummer
    Joined: May 2006
    Posts: 574
    Likes: 83, Points: 28, Legacy Rep: 258
    Location: Finland

    Mikko Brummer Senior Member

    I started with the circular plate, easier to model in CAD. I ran 2 studies with different turbulence modeling: RANS (Reynolds averaged Navier-Stokes) and LES (large eddy simulation), and compared the results against the Potter & Crowther windtunnel experiments as well as Leo's LSP-lifting surface program results.

    The results are good: The LES simulation (orange circles) is spot on the Potter & Crowther presented experimental curve (in black) all the way up to 30 deg angle of attack. The simpler RANS-simulation (green squares) also does well up until CL about 0,8. Above that it falls between Leo's LSP curves for 0% (blue) and 100% leading edge suction (red). Considering the LSP is inviscid and the flow over the plate is almost totally separated, the 100% LE suction model does a great job, too.
     

    Attached Files:

  4. Mikko Brummer
    Joined: May 2006
    Posts: 574
    Likes: 83, Points: 28, Legacy Rep: 258
    Location: Finland

    Mikko Brummer Senior Member

    Flow patterns from the RANS-study, at 10, 20 and 30 deg
     

    Attached Files:

  5. Mikko Brummer
    Joined: May 2006
    Posts: 574
    Likes: 83, Points: 28, Legacy Rep: 258
    Location: Finland

    Mikko Brummer Senior Member

    I only chose the plate because it needed no modelling... It would be nice to do the 2 frisbee models tested, to see if they rate similarly against each other in CFD as in the experiment. Now that you mention it, should have done the frisbee first, don't know when I will have the time for that.
     
  6. daiquiri
    Joined: May 2004
    Posts: 5,371
    Likes: 258, Points: 93, Legacy Rep: 3380
    Location: Italy (Garda Lake) and Croatia (Istria)

    daiquiri Engineering and Design

    Looks nice. However, the flow in the pictures looks symmetric, which in case of a frisbee imo shouldn't be so. Did you include a disc angular velocity in the simulation?
     
  7. Mikko Brummer
    Joined: May 2006
    Posts: 574
    Likes: 83, Points: 28, Legacy Rep: 258
    Location: Finland

    Mikko Brummer Senior Member

    No spin here... the force & surface pressure plots are extensively covered in the AIAA paper for the non rotating discs. While there are some results for the spinning frisbee (fig 8 in the paper), which one is it? Not likely the circular plate. So I took the easy way out. If I do the real frisbee, I could try it with a spin, too.
     
  8. Mikko Brummer
    Joined: May 2006
    Posts: 574
    Likes: 83, Points: 28, Legacy Rep: 258
    Location: Finland

    Mikko Brummer Senior Member

    Animated flow patterns for the LES-simulation. The LES-simulation is transient, time dependent, so in the video you can observe how the flow develops with time, speeded instantly from rest to 20 m/s. The LES also captures more accurately the real behaviour of the turbulence: with LES, the turbulence is resolved directly up to a chosen level, and only the smallest eddies (within the boundary layer) are modeled, while in RANS the turbulence is averaged and modeled all the way. The accuracy comes at a cost, the LES takes about 24 hours for 1 angle of attack, while the RANS runs in 20 minutes.

    In the LES simulation, you can see the smaller vortex cores spinning around the larger, initial trailing vortices, starting from frame 5 or so, while they are absent in RANS. Observe also how the starting vortex for the circular wing is - circular. The animation is about 30 times in slow motion, the whole thing happens in reality in 0,15 sec, the time the air travels through the 3 m long "windtunnel". The trailing vortices are coloured with vorticity, the plate with velocity.
     

    Attached Files:

  9. Leo Lazauskas
    Joined: Jan 2002
    Posts: 2,696
    Likes: 155, Points: 63, Legacy Rep: 2229
    Location: Adelaide, South Australia

    Leo Lazauskas Senior Member

    Splendid work, Mikko!

    Just one minor point - my code is NOT a Vortex Lattice Method (VLM).
    VLM produces rubbish for lifting surfaces with curved leading and/or
    trailing edges because the straight edges of the horseshoe vortices
    "intersect" or come close to the the collocation points.
    That is one reason I like the circular wing as a test case. The other is that
    there is an exact solution for the Lifting Surface Integral Equation.

    I could go into the messy numerical details but it would send you and everybody to sleep! :)

    Leo.
     
  10. DCockey
    Joined: Oct 2009
    Posts: 5,229
    Likes: 634, Points: 113, Legacy Rep: 1485
    Location: Midcoast Maine

    DCockey Senior Member

    I've seen "Vortex Lattice Method" applied to methods with skewed horseshoe vortices which alleviate the problems cited above. A typical application is a tapered and/or swept wing. It's also worth noting that vortex latice methods are equivalent to a panel method which uses constant strength doublet distribution over each panel.

    Leo, a brief description of the method your code would be appreciated by me, though most others may find it a bit esoteric.
     
  11. Petros
    Joined: Oct 2007
    Posts: 2,934
    Likes: 148, Points: 63, Legacy Rep: 1593
    Location: Arlington, WA-USA

    Petros Senior Member

    Very interesting, great work!

    looking at the Cl/Cd polar it appears L/D is only about 4:1? A best L/D is only around 7;1 or so? Seems kind of low from my casual observations. The spinning I think might be a component too large to ignore, greatly affects Rn, which has a very large effect on low speed objects.

    Anyone ever measure the L/D of a real Frisbee? and compare it with these model codes?
     
  12. Mikko Brummer
    Joined: May 2006
    Posts: 574
    Likes: 83, Points: 28, Legacy Rep: 258
    Location: Finland

    Mikko Brummer Senior Member

    Look at the AAIA paper above - they have tested two types of real frisbees and the flat plate I am referring to. Frisbees have a higher lift, but also more drag because they are "thick".
     
  13. Leo Lazauskas
    Joined: Jan 2002
    Posts: 2,696
    Likes: 155, Points: 63, Legacy Rep: 2229
    Location: Adelaide, South Australia

    Leo Lazauskas Senior Member

    There are VLM's that use skewed horsehoe vortices (i.e. a slanted bound
    vortex with parallel trailing vortices) but they are not guaranteed to work
    with curved leading and trailing edge panels. The problem is that a
    collocation point can lie on the (infinite) line through the slanted vortex or
    close to it. This creates very severe numerical problems when the
    influence matrix is inverted.

    The method I use is due to E.O. Tuck and further developed by D.W.F.
    Standingford in his PhD thesis. See the LSP pdfs in previous threads for biblios.

    Essentially, the method solves the once (x-wise) integrated form of the
    Lifting Surface Integral Equation. The method was used by Tuck to solve
    many different types of Cauchy singular integral equations.

    I have also used several numerical tricks we developed over a few years.
    For example, using a Chebyshev grid improves convergence with the
    number of panels in the x-wise and y-wise directions.

    A nice trick is to assume square root behavior at the trailing edge (i.e. the
    Kutta condition) and to build this into the matrix of influence coefficients.
    That increases convergence with respect to the number of panels in the
    x-direction.

    VLM's have trouble capturing the singular behaviour at the leading edge of
    rectangular and curved leading edges. There is often a "kink" in the
    loading near the leading edge. People have developed a variety of
    methods to ameliorate that kink, but the best I have seen is due to
    Standingford. His technique builds into the influence matrix an analytical
    correction that eliminates the kink for rectangular and straight leading
    edges.

    Despite these tricks and techniques, there are still some funny wobbles
    and kinks in the loading near the leading edge of curved planforms. The
    lift coefficient is accurate, but there are still difficulties with the leading
    edge suction. That is a much more difficult quantity to capture accurately.
    CL is pretty easy. Once the leading edge suction is known it can be used
    to calculate the induced drag directly from a balance of the forces on the
    wing. Other (e.g Trefftz plane) methods also work, but I prefer the "direct"
    method for flat wings.

    As I said, I like the circular planform case because it has an "exact"
    solution which makes it an excellent benchmark. The lift slope is
    1.79002303. Peter Jordan used Gegenbauer polynomials to get that value.
    Boersma and others have verified it. Touvia Miloh got close with a
    brilliantly simple approximation, but his solution was for a wing with tiny
    twists near the wing tips, so it is actually the solution for a non-planar
    wing. See the paper I wrote with Tuck for a potted history.

    I could go on and on describing the delights of this wonderful,
    controversial, difficult, classical aerodynamic problem, but it's 5am and I
    must go to bed to dream of singularities and solutions. :)
     
  14. Leo Lazauskas
    Joined: Jan 2002
    Posts: 2,696
    Likes: 155, Points: 63, Legacy Rep: 2229
    Location: Adelaide, South Australia

    Leo Lazauskas Senior Member

    Also, LSP is really only valid for small angles of attack (AoA), typically a
    maximum of about 5 to 10 degrees. The suction analogy I use makes it
    reasonable over a wider range of AoA

    LSP gives zero CL and zero drag at AoA=0.0. An additional component of
    profile drag could be added to give non-zero drag at AoA=0 which might
    improve the correlations at those low AoA.

    Did you use an finite thickness plate in your simulations?
    If so, what is the shape of the leading edge and trailing edge?

    I have seen various results for flat wings with rounded, bevelled, and
    squared-off leading edges. That can make a difference to predictions, but
    it would probably be a pain to grid accurately.
     

  15. Leo Lazauskas
    Joined: Jan 2002
    Posts: 2,696
    Likes: 155, Points: 63, Legacy Rep: 2229
    Location: Adelaide, South Australia

    Leo Lazauskas Senior Member

    There are some advantages for low aspect ratio wings of circular (or near
    circular) planform because of their characteristics at high angles of attack.

    A strange aeroplane dubbed "The Flying Pancake" was investigated
    towards the end of WWII because it could take off from a short runway.
    I think developments in VTOL killed it off. See:
    http://en.wikipedia.org/wiki/Vought_V-173

    These planforms have recently made a come back with small UAV's where,
    as you point out, Reynolds number effects can be very important.
     
Forum posts represent the experience, opinion, and view of individual users. Boat Design Net does not necessarily endorse nor share the view of each individual post.
When making potentially dangerous or financial decisions, always employ and consult appropriate professionals. Your circumstances or experience may be different.