Learning Forum

[FluxLimited]

I am running a transient, incompressible, turbulent (SST k-w), single-phase simulation with the pressure-based solver (coupled solver). I have several different length scales in the problem so I have widely varying mesh densities across the geometry in order to resolve the boundary layer and flow features everywhere. My timestep is dropping to 1e-8 s using the Adaptive Method even with Courant number up to 20. This is unreasonably low. Am I just out of luck with this simulation at my current mesh density?

One thing I note is that for the steady-state solve the residuals take quite a while to drop but when moving to transient solve they drop quickly and converge quickly (albeit with dt = 1e-8). What can this be indicative of?

[aitor.amatriain]

The required time step size is proportional to the mesh size, so what you see is normal.

[DrAmine]

Transient consists of N steps of recurring steady state cycles: for that reason residuals will drop towards the end of time step and increase again at the beginning of the next time step where we start from the previous solution (as if you start every time a new "steady state" run with data coming from previous time stamp).

[FluxLimited]

I understand that, of course. My comment on the residuals is that when running the steady state solve it takes many hundreds of iterations for the continuity residual to drop down to around 1e-2 and they basically stay hovering around there. But if I stop the steady state solve and turn on transient the continuity residual immediately plummets down to 1e-6 in a handful of iterations. In transient I can get the residuals almost arbitrarily low without any problem.

[aitor.amatriain]

The residual of the steady state solver is the residual of the steady state solution, while the residual of the transient solver is the residual of the transient solution. These are different solutions qith different physics involved.
Try to put a time step of 1e-16. Even the most complex problem with a poor mesh converges because essentially nothing happens between the iterations.
In single-phase flows with RANS yo should get a converged solution with the Coupled solver in less than 500 time steps. If not, you should improve your mesh before simulating the transient problem.

[Rob]

Yes, however in the transient solver you need to let it run long enough that the results are changing with the flow field rather than still adjusting from the initial condition. At 1e-6s how may time steps would it take the flow to get from the domain inlet to outlet?