I’m trying to mesh this wall assembly (attached) but it freezes when I try select the parts to apply the meshing settings to. I’ve tried cloud HPC (224 vCPUs, AMD EPYC 7B13) but it still freezes.
I prefer not to ‘simplify’ the model as this way it accurately takes into consideration the number and quantity of thermal bridges present in a given section.
ImprovedCase.pmx (5.0 MB)
Some parts here are extremely thin (0.001 mm ?) and should be meshed with hex elements. There are also small screws. Most parts should be hex meshable, but some would require splitting into subvolumes with 5 or 6 faces (3 or 4 edges each) or modeling as surfaces and using the Thicken Shell Mesh tool. You could try meshing it with tetras like models from your initial threads. But you should first make sure the geometry has no flaws (have a look at that one part with the yellow warning sign) and try removing tiny details such as cutouts and holes. Then there are two ways - compounding everything (if the input and output geometry is correct) and meshing with tetras or (probably better) meshing parts separately and using tie constraints to connect them. Of course, these approaches may also be mixed - you can compound some parts where it works without issues and connect the rest with tie constraints. But simplifying/reducing/defeaturing the model is still strongly advised if you want to get a good mesh and thus reasonable results.
Try to mesh the smallest, largest and thinnest parts first individually. And think about how you will connect all 500 parts later. Maybe merge some of them.
Perhaps you could also utilize symmetry here and simplify the connections between the blocks. Merging parts in CAD software may also work better. Some software has pretty powerful boolean tools, including fuzziness/tolerance and detailed geometry checks to make sure it’s really meshable after importing to PrePoMax. Even FreeCAD has some tools like that.
Maybe you could even consider using submodeling to create a simplified global model and a detailed local model (selected connection), but I haven’t tested it with thermal procedures in CalculiX yet.
I think he probably means the freezing that occurs when trying to select all parts at once to assign the meshing parameters. In this case, it helps to switch the selection methode from Selection by location to Selection by ID.
I would start from much fewer parts, e.g. just one connection:
This shouldn’t freeze or crash and you can easily isolate the issues (especially if you mesh parts one by one).
Surprisingly, the largest parts (plates) can be the most problematic due to their small details inside. As I said, one even shows a warning. It would be good to fix it in CAD first.
ImprovedCase reduced.pmx (3.1 MB)
Exactly – changing to “Selection by ID” helped a lot – thanks.
Thanks @FEAnalyst and @Matej for your insights. I’m still unable to mesh the whole model. I’ve started doing it part by part, and got to this stage with no problems. It really was just a case of re-attaching some part using the Manipulator WB > Align tool in FreeCAD, which is frustrating because it’s ‘random’. Not sure if this is a fault of FreeCAD or PrePoMax.
When adding a third track, it won’t mesh. Have tried re-attaching and re-building some parts (as I’ve done previously) with no luck.
Here are the settings I’ve used. I tried increasing min element size but that lead to errors.
I do get a memory error – not sure if the PC is running out of memory as I’m using a 128 GB RAM remote server:
It looks like I’ll just have to change some geometry, and risk reduced accuracy.
Abaqus can mesh this without preparation if you just use virtual topology on that faulty part and sufficiently reduce the mesh size:
However, a good clean-up is still needed here.
You could also try the built-in Transform tool in the newest versions of FreeCAD. Some small misalignments may indeed break merging/compounding, for instance.
With or without compounding ? In your file, it seems that pretty much only part 45 can’t be meshed, but it has some visible flaws:
Without. I’ve learned to not compound anything unless the mesh works (so as to not waste time). Even when I do compound, it’s only to make it easier to assign materials later on. I always do tie constraints as without it I’ve always ran into errors.
Where about is that flaw and how did you find it?
I’m trying to establish a fool-proof workflow that won’t require me to do trial and error every time I want to mesh something slightly complicated
All the cutouts in that additional row under the third track have that:
It looks like a shell surface or extremely thin-walled surface on the side of the cutout:
Visual inspection revealed it, but built-in checks for small edges/faces and other issues can also be very helpful. FreeCAD has a tool to check the correctness of the parts too.
What’s the tool in FreeCAD called?
Part CheckGeometry, but you could also try the Analysis Situs software for more advanced checks.
Part CheckGeometry didn’t return any errors. Will give Analysis Situs a go.
What’s irritating is that the third track is an exact copy of the first two. So why aren’t there any errors in the frist two? And why does it build the third differently to how it built the first two?
I don’t know how you modeled those cutouts, but if you made a boolean cut with the track then maybe something was wrong with the track or its alignment. As I said, booleans often fail due to small misalignments. Some software includes tolerance/fuzziness settings for them because of that. In FreeCAD, there are fuzzy booleans available in the add-on Defeaturing workbench.
I didn’t model any cut-outs. What I love about PrePoMax is that I can insert parts into other parts and it’ll automatically create cut-outs that usually work.
I’ve tried increasing the Boolean tolerance to 1 mm but that didn’t help either (it won’t accept 2 mm)
I’ve also run several analyses on Analysis Situs with no faults reported
Then I would try recreating them in any CAD software.
This is the BooleanFragments tool, quite specific. You would have to use Part Cut (maybe it will work out of the box) or Fuzzy Cut in the Defeaturing add-on workbench if the regular cut still fails.
There might be no actual geometry errors, such as shell faces, but there are at least extremely thin-walled regions. Some dedicated checks may detect them with proper thresholds.
I’ve finally managed to mesh the entire model. What solved it was:
Changing BooleanFragments > Tolerance to 0.9 mm instead of 0 mm (anything larger led to errors)
Meshing each part separately and then running a manual diagnosis to see what’s causing the error
Compounding parts and then meshing that compound (I’ll still do Tie Constraints when running the model)
I don’t know which exactly of the above solved it, but it only worked after I did all 4. Thanks everyone for your input – it definitely helped.
Here’s an image of the meshed model and here are the .pmx and .step files if anyone wants to have a look.
Something strange is happening when running CalculiX: When I attempt to run it on an HPC server, it fails. Yet when I run it locally on my computer, it works.
I’ve attached both output files. Wondering if anyone ever faced a similar issue
To share with PrePoMax discourse.zip (720 Bytes)
The “Job failed - no results exist” message is a generic CalculiX error. It may mean an incorrect model definition, but if it runs on a different computer, then it’s less likely. Are there any other relevant warnings or errors in the log file from the HPC run (maybe share the whole log file) ? How many elements do you have there ?
