The majority of our CFD simulations are axisymmetric. The advantages of being able to analyze in 2D versus 3D are tremendous for us. Many of our parts are long tubes (40" to 100" long) with wall thicknesses from .031" to .062" and diameters from 1" to 4". Some tubes have a .0004" silver or copper plate on the OD or ID along its entire length. The plating has a significant impact on heat flow but the plating is impractical to model. This is because the plating is so thin compared to the tube wall thickness (77 to 155 times thinner) and because the plating is so long compared to its thickness (100,000 to 250,000 times longer), meshing causes the plating and the tube to have a godzillion elements.
The ability to select an edge and define it as a "surface part" and then to specify its thickness (example: .0004" thick) like we can do in 3D would be a great asset.
Often while checking the final meshcount I have to count the number of digits to make sure if the mesh is of the order of hundred thousands, millions or over 10 million. How about showing the number of elements by separating three digits with a comma? I know it's rather a trivial requiest but can come handy to quickly get the sense of enormity of mesh. There had been times where I have seen my colleagues submitting a job of 23 million mesh, thinking it was a 2.3 million mesh!
Ideally I'd like to have the surface meshing process to finish, even if it could not mesh a particular face, before the simulation aborts. So it could identify all surfaces with (mostly) too rough mesh sizes and put them into a geometry group which can finally be displayed and evaluated.
When we mesh our model, in the message window there is a sentence like this " creating volume mesh" i think loading bar (Percentage bar) must be in it so we can see when the mesh process will finish. Because sometimes meshing process takes too much time.
It is puzzling that while we are meshing, we have to give away a licence in solving, forcing us to stop the jobs running in the queue, just to be able to mesh. In most commercial CFD (or FEA) software packages that I know of, I see that meshing and solving are completely exclusive of each otherin a single licence.
Majority of time of every CFD engineer is occuppied in meshing at his workstation. At the same time, he can't finish his job unless these meshes are run, often in the remote cluster etc. Sorry if I sound rude, but is it me, or has Autodeksk really employed this unfair and unjust tactic to engage a solver licence in meshing, thereby indirectly forcing users to go for additional solver licences, so that they can continue both meshing and solving uninterrupted at the same time?
I know this may call for a major overhaul in code (or maybe not), but is it not reasonable to demand that "meshing" and "solving" be treated exclusive of each other?
We require a good uniform mesh on fans and resistances, which can easily be forgotten by the user, causing poor results or poor convergence.
I tend to recommend 5 elements from inlet to outlet. Could this be automated in any way?
I can appreciate this would be a challenge for irregular resistances but for oblongs, possible?
In Simulation CFD 2013 we introduced mesh history in the design study tree. We're curious how this feature is working for you and what could make it better.
These are some of the suggestions we received prior to having the IdeaStation:
- Be able to add a surface/solid to a mesh history branch. I created the adjustment but forgot a surface/part that I wanted to include there. Drag and drop the branch onto a surface or part would be consistent with some of the other setup tasks.
- Reorder mesh history.. Right click > move up/move down
- When you delete the 'automatic size' level you do a rebuild you get a manual mesh. Interesting consequence, but you eliminate all your history. May want a warning when deleting the automatic size part of the history.
- When you right click on the surface/gap, length scale and go edit. Actually edit those. If you do it to surface it would switch it from false to true.. Length scale would open the diagnostics window.
Do you agree (+ kudos)? Do you have other ideas?
User Experience Designer
I deal with a lot of circular geometries and thus I need to use cylindrical regions to refine small passageways in the model, these can be as small as 0.3mm. Hence I have the main fluid body at roughly 1mm manual mesh, and refine the small gap at 0.1mm. I use a cylindrical region for this.
However I do not need an entire cylinder at 0.1mm, and it would be great if I could refine just a disc shape to alleviate the density of the overall mesh. This is bearable for static runs, but for transient studies with motion as well, the computation time is seemingly infinite.
Rebuilding the mesh definition can be a useful workflow when you want to modify the mesh history and perform sensitivity studies. One issue that slows down this workflow is that if you have mesh refinment turned on before you rebuild the setting is lost.
I think it would reduce the chance of error if when if you rebuild it reuses this setting and when the edge sizing is finished the surface refinement is automatically applied as well so you don't have to click on refine again to eliminate the warning icon.
Could we have some more intellgence in the mesher so that if we have a large model and simply change a wall to a P=0 we are not required to carry out a 100% remesh - should we not have the exact same mesh but without a boundary layer?
Even with manual meshing we require a remesh, seems like time wasted.
Give the option of assigning mesh size scalars to volumes (fine to coarse) before auto-meshing. As a bonus, give the option of bulk assignment of mesh size scalars to material groups.
When doing 2D compressible analysis that include shocks, having adaptation would produce a more efficient and sharper result. Would also be a good test bed to test adaptation strategies before running a larger 3D model.
Certain applications involving both internal or external flow require accurate prediction of flow separation. The mesher and solver should provide new options for predicting the onset of flow separation as accurately as possible.
My CPU is only operating at 3% during the meshing and pre-solve process. Sometimes this takes a long time (up to 2 hrs) on large models. Is there some way you can change the system so the computer uses all of its processing power during this time period.
It will be nice if we can mesh the geometries and check if meshing will be successful before setting up all the material properties, BCs, ICs.,..etc.
It will be nice to know where/how the mesh fails if meshing fails.
I turned on text alerts to let me know when my analysis was complete. The only issue is that the alert comes from a different number every time I get it. Its a simple sequential number but it is pretty irrelevant and comes up as a number I do not know. When I got the first text, I set up the number as a contact titled "CFD Alert". It would be nice if all the alerts came from the same number so I would know what they are just by looking at the contact.
When I change the autosize mesh parameters or the refinement length I find it painful that I always have to click on Autosize or rebuild the mesh definition, remove duplicate autosize entities (which is of course not necessary) and so on, especially if I have some manual refinements, or refinement regions.
All these parameter changes could be automatically inserted into the mesh size definition since the advanced/diagnostics window has to be closed/confirmed with OK anyway.
There are some irritating bugs in the meshing that I would like addressed:
- I use regions to have control on the mesh distribution. But often, change in region size does not have any bearing on mesh adn its no. of elements, and if you keep on changing the region size, you see a sudden change in mesh and element count changes massively. e.g. For region size of 10 until 9.3, there is no change in the mesh, for region size of 9.2 the mesh element count nearly tripples. I have logged a case (08462140) on this earlier in which I was told I had discovered a bug.
- Often, I change the region size on the go with right click on tree and not opening the region dialogue box, changing the size and the "spreading change." But for this to take effect, I have to "generate" the mesh. And whenver I do that, all the materials that are assigned extruded mesh "forget" their setting and switch to being tet mesh. So if you had forgotten to reassign the setting each time you regenerate the mesh, you are in for a disappointment to see the sparse tet mesh spread over the extruded materials.
- The volume growth factor apparently works for fully tet meshes but doesn't work when extruded meshes are present. Even if you first make a tet meh all along adn then try to extrude the surface, with volume factor enabled, it doesn't mesh. I haven't seen this limitation (volume growth factor not available in extrusion meshes) documented anywhere.
- Sometimes, the extruded materials lose the settings just like that, without any reason after refinement, other settings being same! There is no setting that hinders the extrusion mesh like uniform c/s, linear extrusion, 3D, uniform surfaces in the direction of extrusion, ends parallel etc. There is a case (07837801) I have created long back that discusses this anomaly.
- The "approximate element count" is so off the mark most of the times, that it is rather reasonable not to show it until it is withink at least 20% accuracy. For example, in the morning I was staring at the mesh dialogue that predicted the mesh count at 600k while the actual mesh was 4 million. This is more than 600% error!
Additionally, I would really love to see the following suggestions that are probably the limitations and not the bugs considered:
- Currently "uniform" appears in the mesh tree just as a setting, with no indication of what parts it is applied to. At the same time, the extruded mesh setting is not even mentioned. It would be a good addition if both the settigns are shown along with what parts they are assigned to.
- Support for extruded part in all directions. Currently it is possible in only three principal directions.
- Support for mesh adaptation in extrusion meshes. OR at least in tet meshes, in presence of extrusion meshes. Today, the adaptation is prohibited even if one part is extrusion.
- Multicore meshing in presence of extrusion parts. Moreover, making it a bit more efficient. This is a good feature.
- Hex mesh with multiblock capability? I know this is a lot to ask, hence this is last
Sorry for the long list. But I just wrote everything that has bugged me over the year! Meshing is one of the most important aspect of CFD and occupies nearly half of the time of the CFD engineer. I really hope at least few suggestions mehtioned here are considered and implemented.