Flow Simulations, Inc.
Computational Fluid Dynamics (CFD) Models
Provided on a Consulting Basis for Industry
Contours of liquid volume
fraction inside of a crude
vacuum distillation tower
Eulerian multi-phase model

LES Large Eddy Simulation vs. t-RANS

large eddy simulation

We’ve encountered a number of situations where transient flows with large separations occured in industrial equipment. In all of our cases the equipment designs were proprietary and cannot be illustrated publicly. However the equipment had some similar characteristics, causing the flow behavior, so a typical geometry which causes steady state flow problems can be re-created here.

colliding flow iso or side view

colliding flows SS-RANS (steady state with turbulence model),  side view


colliding flows top view

colliding flows SS_RANS,  top view

The model was first run SS-RNG-ke (Steady State, Renormalised k-e turbulence modeling) although this run was a bit problematic due to the unsteady nature of the flow. The model was able to converge 1st order (2nd order in mometum) and results are presented above.

Second, a t-RANS run (transient RNG-ke turbulence model) is being performed.  This is slow due to the near wall discretiztion in order to have manageable wall Y+ values.  Also I wanted to have a reasonable sized model, on the size order of what industrial sized equipment  is.  Also using adaptive time stepping, we have thousands of time steps for a large sized model (over 10 million cells).  Hence results will be a little time in coming in order to capture enough flow time so as to capture any periodicities.

This is just a preliminary animation with an elapsed time of ~ 1sec. It shows contours of Y (upwards) velocity superimposed on an iso surface of velocity magnitude at 25 m/s. Now you know why used contours of turbulent vorticity instead.

Third is the LES.  Of course we haven’t prepared the mesh yet for the LES but will be looking at how coarse we can make the mesh and still capture turblence inducing effects from the walls and eddy sizes.  Some papers discuss that poorly resolved LES may be worse than t-RANS, but since this is a validation case we can always go back and reduce the geometry size in order to get the LES to fit.

Fourth we have Fluent’s DES (Detached Eddy Simulation) which is a hybrid of LES with t-RANS being used for near wall regions.  This is a bit of an over simplification but please forgive me at this time.  There is another option of a “manually controlled” zonal t-RANS / LES hybrid which Fluent has called ELES, for embedded LES.  This approach looks interesting.

Anyhow when I get the the LES portion of the comparison study hopefully I’ll decide on an appropriate model.