[johny_walker]

I had few queries regarding schemes to be chosen while using k-omega sst. I am using this model to simulate aerodynamics flow over a windsor vehicle body in 2D in ANSYS Fluent.



While I select SIMPLE scheme in solution controls, and carry out our simulations, I get a fluctuating wave pattern for the residuals and Cl (please check the attached image. The attached residuals are for 130,000 cells.) which implies that solution is not getting converged. Its the same pattern even if the number of iterations are increased.



However, I was reading on some online forums, it said that I should use COUPLE scheme with pseudo transient condition to get a faster convergence. However, pseudo transient condition should be used only for steady state calculations. Furthermore, I can also use COUPLE scheme without pseudo transient condition but then, I need to select explicit or implicit schemes. About these two schemes, implicit scheme is unconditionally stable and can take higher courant numbers but takes longer time and more computational efforts. Whereas explicit scheme is stable only when CFL<1.

My query is which one to go for? Pseudo transient, Explicit or implicit. If I choose the later two, how do we select the courant's number given that minimum grid size is 0.0008m and velocity is 27m/s.



P.S. The images attached below are obtained using SIMPLE scheme

[vinerm]

The relevance of the schemes is with respect to the stability and not really accuracy. Coupled is better (and you can use it with steady-state as well as transient) only if pressure and velocity have a significant coupling, i.e., they influence each other a lot. But that doesn't mean that you won't get solution with SIMPLE.

We place too much reliance on the residuals, which is not correct. A good part of your solution is that you are monitoring the coefficients for the lift and the drag. If those are fluctuating, then you have to look at the type of fluctuations. At the scale where you have those plotted, it is difficult to say. Use a different scale on the ordinate. If those are stable, the solution is numerically converged, if not, then you need to look at if the oscillations are periodic or not. If those are periodic, then most likely you have to run your simulation as transient and then take a time-average over 3-4 flow-through-times. If those are not periodic, then there is stability issue. First thing to do then is to look at the mesh. Absolute values of mesh size is not important. What matters is wall Reynolds number or . For models, it is recommended to maintain within the order of 1. If that is not the case, you may have to refine the mesh.

[johny_walker]

I do have y+ = 1 on the surface of windsor body and y+=30 on the surface of the ground. Does that effect the results or should I have y+=1 on the ground as well?

[LuckyTran]

You should add to the list which under-relaxation factors to take for SIMPLE because they do the same job as the Courant number in the COUPLED and pseudo-transient solvers.

Just pick one you like. You can play with these schemes forever and optimize them for the rest of your life. It's better to be competent in one than incompetent in many.

Pseudo time-stepping can only be done for steady solvers because time-stepping in a transient simulation would just be time-stepping. On the other hand, a steady solver should not involve an explicit time-step or it would be a transient solver.


Choosing a smaller CFL makes the computational take more iterations and negates the "converges faster property." Btw converges faster means in fewer iterations. It does not necessarily mean they will converge faster in actual cpu time because each iteration can take longer to computer depending on the scheme. You can go for high CFL and converge fast, but your calculation is susceptible to oscillations. Or you can go for low CFL and converge slower, but your calculator is safer from oscillations. URF's work the same way.


SIMPLE vs COUPLED vs pseudo time-stepping really only matters if the P-V coupling is limiting your convergence. If coupling with the energy equation or turbulence is what's limiting convergence, then it won't even matter.