[Bravo]

I am a beginner of SU2. For my research, the flow is low-mach, incompressible flow. However, I found in SU2_AIAA_ASM2013 that it says SU2 only vaild for steady-state only.

Also, this threads, http://www.cfd-online.com/Forums/su2...le-solver.html

And all incompressible test cases officially provided by SU2 developers are all steady cases. Recently, I tried a case, a rigid beam in channel. I can't get the right flow field. The velocity distribution are totally different with ABAQUS/CFD at the same physical time with same material parameters.

So, here, can anybody tell more details about SU2's incompressible solver using Artificial Compressibility method?

This bothers me for a long while.

[Vino]

Hi Bravo,

you can do both steady and unsteady flow simulation of incompressible flows using SU2.

[Bravo]

Hi Vino, thanks for your reply. Can I ask some further questions?

For unsteady incompressible flow, are both the Time stepping and Dual Time stepping good enough?

When I do the unsteady flow, it needs thousands of internal iterations in one external iteration to get a good result. Is this common?

[Vino]

Bravo, its very common to use 1000 internal iterations if you are using explicit time marching, because of CFL limitations. You can go ahead with implicit time marching with very high CFL to get it done in Less that 100 iterations. This is a general idea. explore for your particular applications.

[Bravo]

Originally Posted by Vino

Bravo, its very common to use 1000 internal iterations if you are using explicit time marching, because of CFL limitations. You can go ahead with implicit time marching with very high CFL to get it done in Less that 100 iterations. This is a general idea. explore for your particular applications.

Thank you again! You suggestion is helpful to CFD beginners as me!

[jjalonso]

Originally Posted by Bravo

Thank you again! You suggestion is helpful to CFD beginners as me!

Hi there,

Indeed, as Vino says, SU2 can handle both steady-state and unsteady problems. Let me clarify the content of the discussion a bit further.

If you choose an explicit time step, that means that every cell in the mesh gets advanced forward in time with the same time step, explicitly: no inner iterations required. This time step is the smallest allowable time step in the entire grid (by the CFL condition) which is normally found in the smallest cells in the mesh. For viscous problems, this time step is usually very, very small, which means that, to cover a given amount of physical time (say you want to see what the flow does over a physical time period of 1 second) you may need to do a very large number (in the thousands to tens of thousands or more) of time steps. But this works.

In general, it is more efficient to solve for the unsteady solution using an implicit time-advancement scheme such as the BDF2 (Backwards Difference Formula, 2nd order in time) in SU2. Using BDF2, the advancement of the solution in each physical time step is cast as an implicit solution that requires a number of inner iterations to converge as far as needed (from about 50-100 if you have an inviscid problem with multigrid, to ~1,000 for RANS without multigrid). You have to tweak the parameters of the scheme to get as fast an inner-iteration convergence as possible, to minimize the overall cost of the unsteady simulation. You also have to decide how far to converge each inner iteration:2 orders of magnitude is typically enough.

Good luck!

Juan

[nightserv]

Hi,

I have been using SU2 for a number of flow problems involving incompressible formulation for steady state case, which provided extremely good results. After that, I wanted to try an unsteady simple case (i.e. flow around a cylinder ) in order to get the frequency of the vortex shedding in the wake of the cylinder but with no success whatsoever (i.e. validation).

The boundary conditions are the same, just like in the TestCase (i.e. farfield and no-slip). I am using a physical time of 20s and a time step of 0.05 s with a 200 internal interations/external iteration (dual-time stepping with implicit euler for time advancement) => 400 iterations. The simulation is done for Re =100 using two different flow numerical methods (i.e. JST/ ROE). The methods provided different results in terms of Lift coefficient amplitude, but neither of them is in accordance with benchmark results (approx. cl=0.3). I also use 3 levels of multigrid (V_cycle).

I have attached also 4 pictures with my results.

Therefore, the next question:
1) Is there some kind of another scaling factor?Or am I missing something?or is just not possible to do transient incompressible problems?



Lift JST.png

Drag JST.png

Drag ROE.png

Lift ROE.png

Thank you,

[jiniraj]

How to plot the turbulent intensity in paraview?