6 replies to this topic

[Ninja2b]

Evening all

Im starting work on my Ph.D, doing CFD simulations using Star-CFD. Currently Im meshing using ProStar. Ill be switching to the new Star-CCM+ sometime soon, which apparently gives much more freedom with meshing. Im hoping that I can use SolidWorks to mesh my flows, as ProStar is a bit of a pain to say the least. Other people in my dept. use ICEM or Gambit, so maybe those might be easier to use than Pro-Star.

I was wondering what software the CFD users on this board use, and what the pros and cons of their meshing software are. Any opinions welcome!

Cheers

Cb

[david_martin]

Depends a lot on whether you are meshing in 2d or 3d, what sort of elements you want to use and what the source of your intial geometrical and topological data is.

If you are in 2D, are interested in simple or subparametric trianglur elements, and your initial geometry data can be formulated as PSLGs, you cannot do better than the CMU/Berkley code triangle, by Jonathan Shewchuk. It is probably the most robust unstructured mesh generation code I have ever seen. Great for continuum and Eulerian mesh generation off an initial grid or set of PSLGs

I haven't used ProStar, but I have used Gambit quite a bit - it is quite flexible and reasonably easy to use, but I can't say I like it. Although intended more for continuum mechanical problems than fluids, my favourite commercial mesh generating code is probably Marc Mentat. Its structured mesh generation facilities are astonishing - including hexagonal elements!

Having said all that, I generally work in 2d with unstructured grids and generally use my own code to make my meshes.

[Ninja2b]

Cheers for the info - Ill be working exclusively in 3D so Ill give Marc Mentat a look. Ive found that ProStar is pretty complicated to use, but as Im just starting I dont know if its the same as the rest of the meshers, or just a pain in the ass. I dont like it much, probably because I dont quite understand it yet.

The new CCM+ solver is supposedly able to solve any mesh you can throw at it; unstructured meshes using polyhedral cells should be grand - so Im hoping that will give me a lot more freedom with my meshes. Still waiting for it to be installed though, so I guess we shall see in time!

[CFD Dude]

I've found that while meshing programs aimed at structural meshes (ie HyperMesh and any mesher bundled in with a CAD code, ie I-DEAS) are MUCH nicer to use than the programs aimed at creating CFD meshes (ICEM Hexa, Gambit) it can be difficult to create meshes that meet the quality standards that's required for 3D CFD. I'm not familar with Marc Mentat so it may be a different case, but one needs to be careful when using meshes generated from codes not catering to CFD.

Just because the solver will solve the mesh doesn't necessarily mean the answer is right. Expecially when looking for minute details in the flow (ie Stability analysis, which was the subject of my Master's thesis) poor meshes can have large effects on the results.

[Ninja2b]

Good point about the standards of the mesh... but isnt it typical to run numerous simulations with progressively more refined meshes, until a converged solution is reached, and hence the solution can be said to be mesh - independant? Of course, this solution may not be correct, but thats up to the CFD operator to decide.

I havent used a CAD - bundled mesher yet, so Ill have to give that a go before I take any meaningful conclusions.

[CFD Dude]

You're doing research so it's probably standard practice to check the mesh, but in industry I've seen a lot of people 'guesstimate' a mesh size and accept the results as truth. You can run into a lot of problems doing that.

What I've found most FEA meshers lack that true CFD meshers have is a way of manipulating the cell size near the walls so that the boundary layer gradient is properly captured. This can be done with prisms near the boundary layer or by controlling the cell size to have small tets near the walls and larger elements in the middle of the volume. Either way many FEA meshers don't see this distinction, or it become difficult to set up.

What I have seen tried is to create an outer shell mesh in something like hypermesh and then use TGRID or Gambit to create the volume mesh from the shells. That way you can use the easy interface of the FEA mesher, but have the element control that the CFD program requires. I'm pretty sure you could do something similar with Marc Mentat and ProStar.

[Ninja2b]

I finally got round to trying out Gambit and was very impressed with its functions and how easy it was to use. A colleague and myself spent the guts of a month trying to come up with meshes in Pro-Star and managed to teach ourselves and come up with similar meshes in gambit in around an hour! I was very impressed with the boundary layer functions, where it allows the user to define graduated cell sizes at edges, and increase the size as they move away from the edges - as you describe in your post above. The automesh function was also very good. However, the best feature was the ease with which CAD Data could be imported.

I have talked to some people about ICEM as well, and they seem very happy with it - one of the best features being the way it can patch over imperfect CAD data.