we are running steady state transition modeling with auto timescale (timescale factor=1). however, as we change the timescale factor to 0.5, the transition location changed (as well as the pressure distribution). what's the reason for its change? in my understanding this factor only affect the convergence, but in our cases, we are sure the problem has converged? please, anybody have any comments on it? or any advice, please help.

many thanks

littlelz

Hi,

Have you also done a mesh size sensitivity check? Also checked other tunable parameters (eg convergence residual)? If your simulation is not mesh converged then it does not surprise me you will get timestep sensitivity in a steady state simulation.

Glenn Horrocks

[littlelz]

thanks, glenn

do you mean that the grid indpendence will cause such timescale sensitivity problem? because in my knowledge it is only mean the convergence problems. let's assume a simulation is converged fully, however the mesh is dependent (not refinement enough), will that cause such timescale sensitivity problem? please advice more info

thanks again

littlelz

[Adam S]

glenn is right, if you don´t have done a grid sensitive check (comparison of several resolutions), your solution in fact might be time scale sensitive.

[CycLone]

What level of convergence did you acheive? What are the MAX and RMS residuals?

[littlelz]

max residual 1e-4. is the goal.

in steady state simulation, the goal can't be met in any case
in transient simulation, it can be met easily.

then we compare the results from steady state and transient simulation, there is no difference at all. according to the CFX manual, our understading is the the local separation flow cause this local lack of tight max residual, which will not affect the overall simulation, so steady state simulation will be used

[ghorrocks]

Hi,

Max residuals at 1e-4 is fairly loose convergence.

The turbulence transition model is particularly prone to difficulties in convergence for steady state simulations as the separation point is often moving around and therefore requires a transient simulation for proper convergence. However, as you have found this small movement of the transition point often makes no significant difference to the key outputs (such as lift or drag). Whether this is the case for your model is up to you to determine, you have to do a sensitivity check on your model.

Glenn Horrocks