Enquiry about ibm method

[Robininstry]

Hi dear Xcompact3d developers,

I have some questions about using IBM to resolve cube-like roughness in channel flow. I would like to simulate the turbulence passing by small cube array (only four cubes) mounted on upper and bottom walls. I modified the Case-Channel file so that I can customize the roughness morphology. The simulation is sucessfully done when I use ibm = 1 (direct forcing ibm without reconstruction). But I found that there are pressure oscillation near the cube surface and propagates into the fluid domain. So I try to use ibm = 2. But under identical setup, the simulation diverges at the beginning. I have tried my best to debug, but I failed to run the simulation. I have two main questions. (1) Although the cons of IBM-1 will result in pressure oscillation, is the effect more pronounced for the shape-edge cube case? (2) Do you have any suggestions regarding my case using ibm = 2?

Best regards,
Robin

[mathrack]

Hello,
For the case ibm = 2, have-you tried to slightly update the position of the obstacles ? I think the reconstruction can be difficult if the fluid-solid interface is located very close to a grid point. Otherwise, you can also try to update the accuracy of the location of the fluid-solid interface using the nraf parameter. If you hit some limitations because of the memory requirement, take a look at the following PR : #402

[Robininstry]

Hi,
Thank you for your quick reply. In current setup, the grid node is completely located at the solid-fluid interface to decrease the error caused by ibm = 1. So it this probably the reason for the divergence of simulation using ibm = 2? I will try to reset the position of the cube, however, in case I increase the cube plan area density, the interface will still be close to the grid point. And for ibm = 2, can you provide some recommend number for specific parameters? e.g. nraf, nobjmax, nvol, npif, izap, etc. And I really hope that the pressure oscillation problem will be resolved or relieved by using ibm = 2.

[mathrack]

Hello,

I do not really have a recommendation beyond trial and error... nraf around 10 should do the job but you can increase it to improve the accuracy of the location of the obstacle & the accuracy of the reconstruction. I think you should increase nobjmax if you have a lot of obstacles. nvol is needed only if you want to perform a momentum balance. For the other parameters, I don't know. Check the source code to find out.
EDIT : izap is suposed to skip some points, this might be relevant if the distance between the grid and the obstacles is too small. The parameter is active by default.

Before running the simulation, you can check the minimal value for the distance between the obstacles and the nodes and compare it with the grid size.

As a side remark, if your channel is periodic, you can slightly shift the position of the obstacles in the streamwise / spanwise directions without changing the configuration.