LES, FSI, CFX, Yplus, VIV ... and problem!
3 Attachment(s)
Hi dear cfdonline members!
I'm doing an FSI problem using Ansys CFX. I want to study vortex induced vibration (VIV) of a riser model. The riser model has a length of 9.63 m & a diameter of 0.02m (L/D=482). The velocity of incident current and Reynolds number are 0.84 m/s and 18737 respectively. I want to use LES turbulence model and Yplus should be below 1. It seems that I have to use too fine elements! I've made a mesh. In this case the areaAve value of Yplus is 119.071. And according to formulas of "boundary layer thickness" and "Near Wall Spacing", extracted from "ANSYS CFXSolver Modeling Guide", I have to use elements of order 0.0000185 m. This means a huge number of elements just in the boundary layer!!! This is while I have a rather big domain. I think something is wrong with it. Could you please please please help me solve this problem?!!! 
why are you surprised of the required number of elements? you need to solve the BL so at least 3  4 cells must be within y+<1. You can have a dx+ and dz+ of order 30  40 for LES.
Note that LES is a formulation of the governing equations, not a turbulent model.... 
Yes, wallresolving LES are computationally expensive.
You may want to drop the block structured mesh you are currently using. This enables you to resolve the wallnormal direction better than the other directions, like FMDenaro said. But I would say that X+ and Z+ of 40 seems a bit high. Fröhlich (the only good german book about LES) for example recommends values around 5 for the stream and crossstreamwise direction. Additionally, you might encounter some difficulties when a nonconformal mesh interface lies directly downstream of your obstacle. 
Thanks for your answers. So using LES will impose high CPU demands. What if I just want to study the vibrations (to get the Frequency or amplitude of the vibrations)? Should I solve for the boundary layer?
What about SASSST or DES or other models? I really appreciate your help. 
& one more question! Maybe it sounds crazy, but I want to know if it is possible to use a few layers in spanwise direction (and do a 2D study for each layer!). I know we can't call it a full 3D study, but it could reduce the computational expense.

Quote:

All times are GMT 4. The time now is 09:44. 