I am recently working on a large eddy simulation in film cooling process. I would like to ask you to guidance me in this field. Due to the high number of mesh and small time steps in large eddy simulation, one fluent run lasts more than a few months. It is essential that from the beginning of the run we use final time step (for example 10-5) and for each time step, convergence occurs (for example 10-4 for continuity), or after a RANS run, It is essential that to reinitialize to start solving of large eddy? What has been your solution to reduce the computational time in your numerical work?
Generally, the inital condition in LES is somehow arbitrary and is necessary to run the LES solution for enough time to let the flow physically correlating. As a control, you can monitor the total kinetic energy and wait until the flow reaches a statistically steady state (if there is..). The time required to reach this state can be large but only after that you can cumulate the fields in time to compute the statistics.
Simulating transient eddies (or the approach to steady-state overturn of the fluid adjacent to a solid boundary) has been successfully accomplished with FLUENT. There are many published papers on this subject. As you have discovered, this is a very long and expensive process. Why are you not considering this like any other convective process for which there are empirical correlations? If your purpose is only to accomplish this with FLUENT, what is different about your problem that distinguishes it from solutions that have already been obtained?
Filippo Maria Denaro
Thanks a lot for your response. we need to perform LES for about 10-15 flow-through times, you think that it is necessary to perform LES for this flow-through times? But the most important thing is convergence in each time step before starting to time average the variables. for example we can do LES run in each time step by 20 iteration or less?or we should convergence in each time step (for example 10-4 for continuity)? Prior to time-averaging, can I use time step for example 10-3 instead of 10-5 before starting to time average?
My opinion is not. You need to assess the simulations towards the achivement of a physically correlated state before starting to sample the fields in time. If you let the continuity be not accurate when you increase the accuracy in the residual you simply get a new numerical transient. I don't know the method you are using so I cannot say more.
- Badges
- Science topic