Less likely as it mentions a specific version of Gmsh.
well, that’s quite simple by including CAD geometry (i.e merge “stp/igs/brp”) and mesh size setting (optional i.e general/default coarser) in *.geo files. However, a new thread as suggested can be more clear about the problem.
I’m not sure if it’s a problem I set up or a problem with GMSH. That’s why I used the R -tree algorithm and couldn’t generate a hexahedral mesh
it seems algorithm setup can not be accessed from GUI menus since Hex-Dominant is in experimental states. Needed to set manually in text editor by *,geo files.
thank you ，I will test it
Right, the R-tree (experimental) 3D algorithm is gone from Gmsh GUI in 4.12.2. I hope that they will bring it back improved in the future.
They don’t have a development roadmap but there is some indication at the top of the changelog here:
This is particularly interesting:
new quasi-transfinite 2d meshes (with fans)
Whatever they mean here by “fans”.
Engine_piston.zip (20.2 KB)
I test geo，but it
couldn’t generate a hexahedral mesh
geo：
Mesh.CharacteristicLengthFactor = 0.25;
Mesh.Algorithm = 9;
Mesh.Algorithm3D = 9;
Mesh.Optimize = 1;
Mesh.Recombine3DAll = 1;
Mesh.Smoothing = 0;
Mesh.RemeshAlgorithm = 0;
indeed, i’m also not yet succeed by setting to default coarser and tuning (refined)
however, previous version still can deal with the geometry. It seems my mesh tuning in new version is inappropriate.
*edited
or, maybe the geometry needed to clean up for some extra faces
still did not help, some error message shown by Gmsh logs (v4.12.2). However, i’m not sure and know about the problem.
coreform cubic is an alternative that I would suggest looking at -even with my limited capabilities I have had some success generating quasi-structured meshes some quite complicated structures
indeed Cubit also have similar capabilities in Hex-Dominant meshing. Is there any improvement in volume fraction & times since implemented to the latest versions?
Also, is possible to share a mesh file generated with finer ones, similar density and type (linear)? Thank you.
sreenshot from Tautges et al. (1998)
Problem with Cubit is that is not free unless you are a US defence supplyer
Salome_Meca also has nice hex meshing capabilities but they seem to require a lot of manual work (especially the hexa block technique) and there are some serious limitations. All in all, Gmsh is probably the best free hex mesher.
But if new versions of Gmsh can’t handle hex meshes as well as older ones then I think that adding an issue in their GitHub repository would be a good idea. Even if they are aware of this regression.
HalfPiston.inp (4.0 MB)
I’m not sure about the quality of the mesh or how to measure it. I have attached a slightly finer mesh for you to look at. The non commercial version of cubic has a limit on how many elements it will export so I generated a half model.
1 Like
it seems not hex-dominant in use, element statistic shown full hexahedral element. Also, if i can understand properly, coloring segment in your screenshot indicated some partition is defined. But thanks anyway, still interesting in their Hex-Tet algorithm and results.
It may be possible to split the piston into a quarter model - and then use some transfinite meshing.
Just my 2 cents on this “beatifull” hexa meshes. Why bother trying to get an hexa mesh with limited tools as Salome or Gmesh, in a model that will not be really usefull? I meant, normally when we do hexa mesh is to get accurate stress with few elements/nodes, but all the solid models that I see in this thread lack of some radius were there would be high stress hot spots, but due to the lack of radius that were removed to get an hexa mashable geometry those models are inaccurate… because we lost the focus on the problem, get the most accurate stress, that most of the times happen in this internal radius that an automatic second order tet mesh would capture very easily. And if is for a thermal analysis…cheap lineal tets are enough as well.
I have spent half of my carer doing hexa mesh on complex geometries, but using comercial tools as I-DEAS, NX or Abaqus, but we did it with all the radius in the stress concentration zones, and we use hexa because they were rubber parts that suffer a lot (horrible) deformation and contact.
On these days with such a powerfull hardware and multicore solvers, is cheaper just to use automatic tet with some refining than taking a lot of time to get an hexa mesh that cannot get the stress were it neededs. Maybe were I would justify hexa meshing will be in parts with hyperelastic/foam materials under very high deformations, forging or impact simulations, but not for a lineal static analysis on a steel/cast/alum part.
4 Likes
I agree 100% with you. I’ve analyzed similar models with hexas and tetras, and the differences are minimal if you have adequate modeling practices. That’s where the experience really comes from, and clearly you have it 
!
1 Like
indeed, me and anyone also maybe seeking automatic generate hexahedral mesh as tetrahedral available i.e no partition. Hex-dominant seems potential, even not yet ready and perfect.
below simple example, it seems model B have balance between computational resource and result accuracy when some MPC needed also can generate automatically.
model A, Gmsh v4.x (C3D8I,C3D4) filler 1st order tetrahedral
model B, Gmsh v3.x (C3D20R,C3D10) no MPC automatically defined
model C, Netgen (C3D10) large number of nodes, 1st order not usable
Have you tried with the C3D8I?
Maybe someone remeber ALGOR Simulation? It had a good algorithm for mixed mesh (hexa, tetra, wedge).
More than 10 years ago Autodesk purchased ALGOR and all the rights but the project I believe has been abandoned. If I’m not mistaken, now Autodesk FEMs only use meshes with tetrahedra