Update: I’ve been discussing this with the HPC service provider and they pointed out that changing the solver to anything other than Pardiso solves it. So perhaps it is a compatibility issue after all? Maybe it doesn’t work as it should with Windows Server 2025?
I like to do everything within PrePoMax rather than having to start a CalculiX instance.
Ok, I see. But then it’s a problem on the solver’s side. If it’s confirmed and discussed on the CalculiX forum, it might be good to report it on their GitHub: Issues · Dhondtguido/CalculiX · GitHub
The solver works fine if it’s run standalone (not within PrePoMax). It is only when I run the solver on PrePoMax on the HPC that I encounter problems. It runs fine within PrePoMax on my laptop – that’s why I suspect it’s a compatability issue
All right, PrePoMax may indeed have some compatibility issues with this system since it’s meant for regular Windows versions. I’d rather expect some graphics issues, crashes and so on, though. It’s also strange that it works with other solvers. And that the error message comes from the solver.
This is the first time I’ve seen someone model such an advanced model. They never do these for construction analysis. It looks like it’s being designed for thermal calculations. Your simplification of the brackets will result in significantly inaccurate results. Are you also adding anchors? In my opinion, it would be better to model a much smaller range. A suspended ceiling made of solid panels? Your model suggests they should be composite panels with thin 4mm walls. With 50mm thick solid panels, it’s a lot of weight for aluminum profiles and a lot of fasteners. Expensive. A huge calculation model. I’d love to see a snippet of the result 
Yeah, at the beginning of this thread and maybe also previous threads from the OP, we pointed out the large size and geometry complexity of this model. Usually, one should find a way to simplify it. Model only a selected region (e.g. connection), utilize symmetry, assume more basic shapes of some complicated components and so on. This would speed up meshing and solving significantly. Even if the model is eventually solvable, refining its mesh (to ensure accurate results) may become impossible.
So I know how to address this, are you a construction/architecture professional? This kind of analysis must be done if you want to compare the thermal performance & cost feasibility of various facade support systems.
Which bracket? Have you tested this, or are you making things up? How do you know it’s a simplification? Have you seen the end product?
I’d avoid commenting on things I’m clueless about.
Again, you’re showing how clueless you are about the model. If you knew anything about construction, you’d know that a slab is a horizontal element, and what you thought was suspended ceiling is natural stone cladding. Seems like you had it upside down
Please share with us the feasibility study that you’ve done that proves there’s a better and cheaper alternative.
I don’t question how things should be done in fields I’m not an expert in and I expect any sensible human being to do the same. I hope you refrain from commenting on things you’re clueless about so don’t have to waste my time responding to nonsense.
In principle, yes. The reason why I modelled a whole external wall section is so I can take into account the heat loss via an uninsulated/insulated floor slab, and the ratio that that heat loss represents out of an entire wall section (i.e., full height). If I only model e.g., one stone panel then that ratio would be skewed. The end goal is to take the heat loss per m2 of external wall area so I can determine total heat loss, cost, etc.
@CosmoKramer I see. How long does it currently take to solve when it runs in the best case ? Do you think it’s possible to further simplify/reduce it in a different way ? I’m not familiar with such civil engineering topics, but there are often ways to implement some multiscale approach (including 1D/system modeling) or to make various simplifying assumptions. Perhaps you could also use submodeling here to focus more on some regions of interest.
In the order of time consumed: tie constraints (4 hours, HPC server) > meshing (15 mins, HPC server) > CalculiX (10 mins, laptop). The reason why I do last one with laptop is because I can and because there are still some issues with running PrePoMax on Windows Server.
I’m not sure if there’s a way to further simplify it. I always start with most complex and then simplify if there are issues. So this is the most complex version that actually runs without errors. I don’t see the need to simplify further as it works.
The most irritating part of the whole process is meshing. I’ve probably run hundreds of meshing processes and I really do think there’s something irrational happening. Many times alI I had to do was to literally delete and re-create the part and that solved the problem. Not sure if that’s on FreeCAD or PrePoMax. But I’m making notes on best practices so I can get it right first time when I’m modelling something similar.
One thing I meant to ask is why the heat flow isn’t showing as negative, when the direction of heat flow should be in the negative y direction:
The sign convention for RFL is the same as for loads (flux entering the body is positive while flux leaving the body is negative). It’s different for Section print, though. And RFL can be confusing because it also includes input fluxes.
Could you elaborate – what’s section print?
I opted for RFL because it roughly matched my hand calculations when I did a simple simulation – FLUX was way off.
It’s a newly implement type of history output, based on CalculiX’s *SECTION PRINT keyword. It provides integrated heat flux results from the selected surface. Here the convention is such that positive value means flux leaving the element through the surface.
This measure is fine, but you just have to watch out for regions with applied flux. You may need to separate them from RFL measurement regions. Here’smore about it: Accuracy of Reaction Flux Plot
So in my case would section print be automatically assumed? In that case the positive sign would make sense – i.e., heat leaving the surface.
I forgot that the section print feature in PrePoMax doesn’t support heat flux yet (only forces and moments). But you can add it manually via Keyword Editor:
*Section print, Surface=surface_name, Name=SP-1
SOF
Then you will find the result in the .dat file from the analysis (can be found in the work directory):
How do you know what I’m doing? I can’t see the end result, but I notice you’ve changed the pressure on the aluminum brackets from profiled to flat. I also immediately noticed what you’re doing, for example, that you have 20 cm of thermal insulation. But you’re right, I don’t need to know what you’re doing. You’re creating a very accurate model, which I think could be simplified. I’m not asking you to calculate for a tiny section, but I think 1/4 of the volume is enough. The PN-EN ISO 10211 standard doesn’t require such a large section. For stone, I’d choose stronger brackets and slightly fewer fasteners. Models of this size are very rarely made, even for high-rise buildings. This is because flat facades aren’t made and no one has that much time. There are so many details, and your fragment is trivial for the contractor, but difficult to calculate. What will you do if you also have an aluminum curtain wall with insulating glass? That’s my advice. Such an accurate model is made to assess water condensation. For heat loss, such precision is only required in research, which suggests you have access to a computing cluster. I’d really like to see the result. My description isn’t meant to be malicious, but I’m following your work out of curiosity. I applaud your persistence and diligence. You don’t have to respond. If you felt offended, I apologize.
FreeCAD (or its geometric kernel) may sometimes fail to generate correct geometry. Netgen or Gmsh meshers in PrePoMax may fail too - this happens mostly for models with geometric flaws or very small features. PrePoMax has some tools that may help (e.g. defeaturing and CAD geometry analysis), but most work has to be done in CAD software before importing the geometry for analysis anyway.
The problem with meshing is also that even if it works, very small features along with large parts may result in lots of elements. Of course, it’s good to mesh anything applicable with hex elements (also to reduce the element count), but in your case, it’s rarely possible because of all the connection details.
Another tip is that it’s better to start from a coarser mesh, see if it generates and keep refining if needed. Sometimes you may have to go really low with the maximum element size due to those small details. But if you start from a coarser mesh, there’s a chance it will work with fewer elements. Then you can always refine (possibly only locally) to ensure correct (converged) results.
I’ve noticed that it’s worth planning the mesh density right away by additionally slicing the model, as you’re doing. This allows for excellent mesh control. Therefore, I would cut cylinders around the steel anchors to accommodate the transition mesh..
Yes, partitioning may help too. It’s manual and sometimes tedious, but doable in FreeCAD. There’s a whole chapter about it here: https://wiki.freecad.org/FEM_Geometry_Preparation_and_Meshing#Geometry_partitioning
It helps obtain more regular mesh and allows much more control over local refinement, among others.
I never made assumption about what you do – that’s why I asked what you do to see where to start in responding to your…uninformed comments. And even if you were on construction, it appears there are many things you don’t know about (which is fine – just don’t assume that you do)
Well actually you do, if you’re going to comment about the way I’m doing it.
Enough for what? You don’t know the end goal here – you never asked before going on about the best way to do it. What’s EN 10211 got to do with anything?
Again with the uninformed assumptions. I asked you to show us the study you’ve done do reach that conclusion, and so far you’ve failed to do so.
They are often made if you want to compare the feasibility (cost vs. thermal performance) of various cladding support systems. They are also made by any manufacturer who wants to showcase their system. Or when the facade’s thermal performance is critical – e.g., Passivhaus projects.
I’m not offended – I take issue with people making assumptions about things they don’t know/don’t understand. And you’ve done it again in your second comment. I really hope you stop doing that, even for your own sake.
