Setting Up and Monitoring Transient CFX Simulation

Jack001 · April 28, 2016, 11:04

Hello,

I have a steady RANS simulation of a case I want to run transient LES on. I tried running the transient simulation from scratch, but the solver crashes in the beginning. I have read that it is instead recommended to use the RANS simulation as an initial starting point for the LES. I am not sure how I go about doing this in CFX however! Does anyone have experience with this?

Secondly I am wondering if it is acceptable to run my LES on a very very coarse grid and larger than desired time steps (so that I can actually get results on my desktop computer) before I attempt to run a proper simulation on a remote cluster. I am very new to this area so I just want someone with experience to chime in - is this an acceptable way to do it? Otherwise one would have to run the full LES simulation and it would take days/weeks/ even months before one gets a solution to verify the physics, and obviously that is not sustainable as there are bound to be mistakes! So i guess im wondering what is a good way to monitor the integrity of the LES simulation so that one doesnt have to wait ages before finding a mistake!

Thanks!

ghorrocks · April 28, 2016, 22:07

It is not good that your LES model crashes when you start. That suggests something is fundamentally wrong and is likely to cause problems even for the run which starts with RANS. I would investigate this further as if you fix this it is likely all your runs will run a lot better.

You should always run simulations with coarse meshes first to check it is functioning as expected before you do a fine mesh run for accuracy. This is standard practice for CFD simulations.

Jack001 · April 29, 2016, 14:01

Thanks for the help Glenn.

I did some corrections and now my case runs at least. However I am wondering:

Is the reason to use a converged solution steady as starting point, that the system is not "shocked" to much by a non-physical initial guess, which could lead the unsteady solution to diverge?

Currently my simulation runs for a 1 or two time steps (with max 10 inner iterations) but then encounters a

c_fpx_handler: Floating point exception: Overflow

Could this be because of what I mentioned?

Thanks!

ghorrocks · April 30, 2016, 10:44

In some, but not all circumstances a non-physical initial condition can cause problems. If the simulation is very stable it can handle it and quickly pulls it back to reality. If not stable - well then it will crash.

There is a FAQ on floating point error which discusses a few things to consider (http://www.cfd-online.com/Wiki/Ansys...do_about_it.3F) but you are correct in that a non-physical initial condition may be causing the problem in your case.

April 28, 2016, 11:04	Setting Up and Monitoring Transient CFX Simulation	#1
Jack001 Member Jack Join Date: May 2015 Posts: 98 Rep Power: 11	Hello, I have a steady RANS simulation of a case I want to run transient LES on. I tried running the transient simulation from scratch, but the solver crashes in the beginning. I have read that it is instead recommended to use the RANS simulation as an initial starting point for the LES. I am not sure how I go about doing this in CFX however! Does anyone have experience with this? Secondly I am wondering if it is acceptable to run my LES on a very very coarse grid and larger than desired time steps (so that I can actually get results on my desktop computer) before I attempt to run a proper simulation on a remote cluster. I am very new to this area so I just want someone with experience to chime in - is this an acceptable way to do it? Otherwise one would have to run the full LES simulation and it would take days/weeks/ even months before one gets a solution to verify the physics, and obviously that is not sustainable as there are bound to be mistakes! So i guess im wondering what is a good way to monitor the integrity of the LES simulation so that one doesnt have to wait ages before finding a mistake! Thanks!

April 28, 2016, 22:07		#2
ghorrocks Super Moderator Glenn Horrocks Join Date: Mar 2009 Location: Sydney, Australia Posts: 17,841 Rep Power: 144	It is not good that your LES model crashes when you start. That suggests something is fundamentally wrong and is likely to cause problems even for the run which starts with RANS. I would investigate this further as if you fix this it is likely all your runs will run a lot better. You should always run simulations with coarse meshes first to check it is functioning as expected before you do a fine mesh run for accuracy. This is standard practice for CFD simulations.

April 29, 2016, 14:01		#3
Jack001 Member Jack Join Date: May 2015 Posts: 98 Rep Power: 11	Thanks for the help Glenn. I did some corrections and now my case runs at least. However I am wondering: Is the reason to use a converged solution steady as starting point, that the system is not "shocked" to much by a non-physical initial guess, which could lead the unsteady solution to diverge? Currently my simulation runs for a 1 or two time steps (with max 10 inner iterations) but then encounters a c_fpx_handler: Floating point exception: Overflow Could this be because of what I mentioned? Thanks!

April 30, 2016, 10:44		#4
ghorrocks Super Moderator Glenn Horrocks Join Date: Mar 2009 Location: Sydney, Australia Posts: 17,841 Rep Power: 144	In some, but not all circumstances a non-physical initial condition can cause problems. If the simulation is very stable it can handle it and quickly pulls it back to reality. If not stable - well then it will crash. There is a FAQ on floating point error which discusses a few things to consider (http://www.cfd-online.com/Wiki/Ansys...do_about_it.3F) but you are correct in that a non-physical initial condition may be causing the problem in your case.