https://linen.dev logo
Join DiscordCommunities
Powered by
# les-ras
  • c
    It would be great
  • z
    or perhaps one on turbulence that will be about all the turbulence models
  • l
    yes the thing is that there are 100 possible hashtags, but I like the idea of moderators per channel @User
  • s
    > yes the thing is that there are 100 possible hashtags, but I like the idea of moderators per channel @User @User No problem, glad to help from time to time
  • s
    > or perhaps one on turbulence that will be about all the turbulence models @User For me no problem, I will moderate on LES related topichs however
  • t
    Has anyone a good source or experience with the cyclic boundary condition and the space needed between inlet and outlet for LES?
  • s
    On channel flows you mean?
  • s
    wall-resolved?
  • a
    In general if you have a channel half height of H then you will need atleast 7H in the streamwise. Remember however if you use periodic bc on the spanwise, the length scales are not constrained and you will have to think a little bit about the resolution in spanwise. Having said that something like 3H or 3.5H should be enough in the spanwise.
  • a
    Highly recommend Moin and Kim, 1982 paper.
  • s
    > Highly recommend Moin and Kim, 1982 paper. @User the choice of 2\pi H \approx 7H is partly due to the fact they use FFT in the horizontal planes. They showed that distance to be enough to guarantee periodicity not to produce correlated fields along the streamwise direction.
  • a
    @slopezcastano indeed, we did some DNS tests (using OpenFOAM) and found that 7H for channel flow is gives pretty decent results for 7H. Just curious what would be a good length in your opinion if not 7H.
  • s
    Depends on what you want to see. I got perfectly reasonable results with 4/3\pi x 2/3 pi x 2
  • s
    depends on what you want to see, and the Re_tau you want to go
  • a
    To be exact, we had 7.5HxHx3.12H streamwise, vertical and spanwise respectively for Re_tau=250,500,600
  • s
    We did for re_tau = 800
  • a
    @slopezcastano oh, interesting I'll give this a go and see if I can reproduce the results. Thanks!
  • s
    both DNS uDNS and LES
  • a
    Did you publish these results? Would love to read that! Thanks
  • s
  • a
    Thanks
  • t
    Thanks for the Input! I recently found some paper, where they did a LES of a ripped pipe wiht only on rib, here a example: https://web.stanford.edu/group/ctr/Summer/SP14/09_Large-eddy_simulation/10_zhu.pdf The author also showed in this PHD thesis that the influence on the pressure loss, which I am intrested in, is nearly the same for a longer streamwise geometry. So I am wondering because I thought that should not work.
  • a
    As @slopezcastano suggests, you could try a smaller domain of (4/3)*pi x (2/3)*pi x 2 for a CHANNEL FLOW (not sure if it translates to a pipe). You can then check the two point correlation in space (Rxx-streamwise,Rzz-spanwise) to confirm that the length is indeed large enough.
  • s
    In a pipe you have axisymmetry so you need only think of streamwise and radial coordinates. I think coherent structures are invariant for channel/pipe flows
  • s
    Two point correlations may surely help in clearing such issues
  • a
    Out of curiosity, what kind of resolution do you generally use for the spanwise direction @slopezcastano ?
  • s
    for wall-resolved LESin channels? dZ+~16-25
  • a
    Cool! Thanks
  • s
    you need at least 4 cells to represent a vortex, and the size of turbulence generating horshoe vortices is around dZ+=100
  • s
    *horseshoe
1...789...52Latest