I have to admit that, unfortunately, I don’t quite understand exactly what you want. Is it:
Do you have a component or an enclosed space, and want to optimize the component so that it fits into the design space and is as light as possible? Then you need a shape or topological optimization method. The on referenced by stoli looks promising.
Do you have a component that is mostly properly dimensioned but has local stress peaks that you want to reduce? Take a look at the CAO method in: C. Mattheck, Design in Nature
You’ve performed an optimization and want to convert the result (which is probably closest to an STL file) back into a CAD model? That’s probably the most difficult challenge. I’m not aware of any way to automate this process. It’s the same issue as converting 3D scans into CAD parts. It probably won’t be possible to avoid manual reverse engineering.
Matthek triangle method is a potential smooth algorithm to extract new good surfaces after topology optimization in beso or other tools.
Also levien splines can be attract to the outer nodes by beso result model.
At the end of such a module you can make a new solid by surfaces and then mesh this, calculate this and perhaps you go to end or again in beso,
second: It is possible years or decades to implement both splines in CGX or other calculix postprozessors to produce good models with good bionic curves instead of edges or circle curves.
so often you do not need BESO or other optimizers with a good first model.
Third: You are free to have a more genius idea to make a module in beso or calculix or anywhere to a very good or best cad-model after beso result.
I actually wrote a script a long time ago that generates smooth CAD surfaces (NURBS) from STL files. See the screenshot: on the left, the Blender Monkey “Suzanne” in Blender as an STL file, and on the right, in a CAD program as a NURBS surface model.
Unfortunately, I never got the script to the point where I could release it to the public. It probably wouldn’t work with the current version of Blender either. It wasn’t very efficient, and could only convert a few hundred polygons; otherwise, the conversion took several hours. Plus, the mesh had to be of a certain quality (as the saying goes: garbage in, garbage out).
If your goal is simply to (re)mesh an STL file generated by an optimization simulation in order to do another simulation with it, you might also want to take a look at the GMSH createTopology feature:
@Nobody-86: the result on the "monkey Suzanne picture is very impressive; I’m wondering if the work is based on the litterature (if so, do you accept to share the sources?), or on your own developements?
Thank you very much for the compliment. I didn’t use any specific literature, just my own ideas and the manuals for Blender and gmsh (which I use as a background process for the conversion).
Well, now I want to add a couple of things to this topic. The post processing of optimized solutions is a research topic on it’s own. My work is focusing the topology optimization itself. On the other hand I started to develop a couple of things, to make comparissons of different designs possible.
The workflow is self-thought and does not relay on any literature whatsoever. I tried to get rid of the most likely errors etc.
Here is a example (middle picture) of a optimized design. When using the SIMP method, you’ll get “red” areas / faces, which correspond to non-physical density equal 1, and the other surfaces, are e.g. a threshold of 0.5 for the blue faces. The red ones appear in areas, which are excluded from the topology optimization and faces, which are equal to the outer boundary of the design domain. These element faces are “fixed” and therefore shouldn’t be processes further. All other faces should be smoothed via an smoothing algorithm. Several are available.
Now my example:
First all “red” areas are isolated. The others are smoothed via the “Taubin smooth” algorithm. And these meshes are reconnected afterwards. Still there are several steps necessary, to get a high-quality FE mesh, but for a first evaluation the result is good enough. The Arrows indicate those fixed areas. The others are smoothed with the proposed algorithm. The red lines show the edges.
After all, the smoothed areas should be optimized via a shape optimization, to reduce the stresses and enhance fatigue life. This is all free-form. Mattheks approach is an algorithm to reduce stresses in conventional designed parts, with high stress concentrations in radii, which are placed by drafts men in very bad locations.
A better characterization of his work (IMO, of course) would be to see it as a generalization of a shape optimization found in nature in different forms.
This is ineed a far better explanation. I can confirm this.
Something to be added:
Here is a quite impressive industrial application (from the TOSCA user manual), which shows the potential in stress relief, when using optimization. This was in fact a shape optimization. The results were later simplified to have three connected radii for conventional machining instead of free-form. In the end, it is quite similar to Mattheck’s approach.
The part you are showing does not look like a machined part?
It looks more like a cast part. It could also be closed die forged, but one would expect a forging to be stronger.
That a break occurs where it does is not really a surprise.
The transition between the solid part and the hollowed out part just looks wrong to me.
You are right, the part is not machined. In fact it is manufactured via aluminum die casting. So the actual die is manufactured via machining. A free-form transition in the high stressed area (which is a result of a shape optimization) would lead to high cost within the die. Additionally in the realm of conventional machining, it is all based on defined radii / starting / ending points. This is necessary, for quality assurance, CAD, CAM etc.
The bionic / organic structures from topology optimization and also shape optimization are therefore a nightmare for conventional machining. The method by Mattheck is a good example for bringing free form and conventional machining closer together, via simple geometrical structures, which are close to bionic ones.
Something to add to this automotive bracket (Audi AG). Most engineers would say, this looks odd and will brake. My theory is, that the drafts men had a mass-target and didn’t optimize with existing tools or even analyze (FEA) what so ever.
