[Crank-Shaft]

Hello CFD Online,

I am currently working on a project which is extremely resource and time-limited. This involves a relatively simple geometry - flow around a 3D, prismatic, circular cylinder - as shown in the image attached.

However, the problem is that I am currently having to block and mesh not only based on the geometry definition but also to ensure that all the blocks have an approximately uniform number of cells. This is required since we are trying to maximise the efficiency of our runs on the small cluster (4 nodes, 4 procs each hyper-threaded) and the in-house code needs the total number of blocks > total number of processors.

So I have tried this 4 times already and the current mesh was generated using 24 blocks (O-grid+tail+midplane split). I desperately need a better way to split around the cylinder and the O-grid without affecting the existing mesh. Is it possible in ICEM to split along the existing cell interfaces by nodal locations? This would make it a lot easier and I would be able to preserve the node distribution.

I would like to split the existing blocks as shown in the PNG drawing - DomainProposedBlocking.jpg
Is there an easier way to get approximately uniform number of cells for each block?

Mesh Files for ICEM

I look forward to your response and seriously, urgently need some ideas on this. Please share your thoughts if you've encountered any similar problems in the past.

Thank you!

[Crank-Shaft]

Hey everyone,

I still haven't received any suggestions for the issues mentioned above, however, I thought I'd post some updates for those interested.

I managed to split the central O-grid around the cylinder into 16 blocks instead
of 8 blocks by using index control and then using a vertical and horizontal plane for the slice. This current state is shown in the image attached.

This has a total of 28 blocks and the node distribution per block is also shown
in the attached image. Unfortunately, this is still imbalanced and I really need around 60-80 blocks to run on the larger HPC for an efficient parallel simulation run.

Does anyone know how to preserve the bias and distribution of the 160 radial nodes around the cylinder O-grid, but split these blocks further using a concentric, circular cut?

I desperately need to split the 160 into 3 lengths of 50,50,60 grid points each. This would give me a total of 160 blocks for the entire domain, which would be perfect!

Hope this makes sense and I would greatly appreciate your response.