[cons013]

Hello there,

I am investigating ways of speeding up an unsteady run but without compromising too much on accuracy. The current settings I am using seem to be very promising, and I was using the standard deviation and variance of the data set (in this case, for lift and drag values) to help drive any decision.

One of the ways I sped up the simulations was increasing the timestep and implementing less relaxation, among some other settings. I found that one configuration resulted in a lower std. dev and variance of the lift values than the benchmark case (currently used settings in our workflow).

What I want to know is, is this a useful metric at all? I worry that by using a larger time step, there may be some smaller scale flow effects that are smoothed/destroyed, which falsely improves these metrics. Any other valuable ways of judging the quality of an unsteady sim would be much appreciated!

[FMDenaro]

Originally Posted by cons013

Hello there,

I am investigating ways of speeding up an unsteady run but without compromising too much on accuracy. The current settings I am using seem to be very promising, and I was using the standard deviation and variance of the data set (in this case, for lift and drag values) to help drive any decision.

One of the ways I sped up the simulations was increasing the timestep and implementing less relaxation, among some other settings. I found that one configuration resulted in a lower std. dev and variance of the lift values than the benchmark case (currently used settings in our workflow).

What I want to know is, is this a useful metric at all? I worry that by using a larger time step, there may be some smaller scale flow effects that are smoothed/destroyed, which falsely improves these metrics. Any other valuable ways of judging the quality of an unsteady sim would be much appreciated!

Without full details is impossible to answer. Formulation, accuracy order, grid size and so on …

[cons013]

Originally Posted by FMDenaro

Without full details is impossible to answer. Formulation, accuracy order, grid size and so on …

Hello,

This is using the GEKO-based SBES solver in fluent, with bounded CD discretisation. The grid size is on the order of 1e8 with a timestep of the order 1e-4. I can't give many more details due to NDA, but hopefully that can help you get an idea.

[FMDenaro]

Originally Posted by cons013

Hello,

This is using the GEKO-based SBES solver in fluent, with bounded CD discretisation. The grid size is on the order of 1e8 with a timestep of the order 1e-4. I can't give many more details due to NDA, but hopefully that can help you get an idea.

Time-step in RANS formulation has no real physical meaning owing to the statistical solution. Time accuracy is not an real issue.


More relevant are some specific parameters in GEKO formulation as detailed here
https://www.ansys.com/content/dam/am...-Ansys-CFD.pdf