Topological Optimization

jacob

Yes, the whole SIMP is written in python in separate script.
Stamping and casting seems to work, so I just want to finish the symmetry.
Sooner or later I'll upload it to github and create some manual.

mf

What do you mean by casting? Do you want to do topology optimization of runners (this is a CFD problem)?

jacob

mf you do casting simulation with CA?

mf no, I am just trying to create some general opensource software for structural topology optimization, which can deal with more complicated boundary conditions than some basic and more than one design space with different manufacturing controls with code_aster as an fea solver for statics.

jacob

https://paulino.princeton.edu/journal_papers/2016/AES_16_TopoloyOptimizationWithManufacturing.pdf

Runners?

mf

jacob
Runners are the lines that connect your part to a casting machine in casting. They can be optimized with topo optimization, but a CFD software is needed. See example below (it's not a good example, because it has no vents and other casting rules are violated).

I understand now. You want to topo optimize a part, that can be cast later (it has to have draft angles and must be releasable from the mold).

Sound_Spinning

mf
I don't think @jacob is talking about the casting process itself.
In TO you can have a 'casting' structural constraint, which means the resulting structure can be made by casting. Not to do with fluids during casting, that's moldflow kind of stuff, isn't it?

mf

Sound_Spinning
Yes, I realized that. But when you say 'casting' to a person that has done casting simulation for years, then this is what this person will think of.

It's nice to do that in TO too, but in an industrial setting you will have to redesign for castability, anyway.

Sound_Spinning

mf
Separate things, you shouldn't optimise the flow before you do the weight & stiffness of the part, which is what structural TO does 1st, mainly at concept stage of design. Then you can still get ways around that on the moldflow part, but with a very good starting point from the structural performance 1st.
I've done a lot of structural TO in the past in production, and if done half right at the start, I didn't have a problem with those designs with the 'flow' team. Both metals & plastic injection moulding stuff.
But I get you, it all has its quirks, it's about finding a balance across so many CAE tools/teams.

mf

Sound_Spinning
Sure, the cast part should be finished before flow sim. Also, having influence on the cast part or not depends on if you are a supplier of your simulation work for casting or not. If I supply the optimization of the filling of the cast part, then my influence on the part itself is limited (in reality near zero, because the customer would have the part optimization on his own or before the cast part will be designed; in practice, this is just adding drafts and other casting specific stuff). Also, I have never seen mechanical TO on a cast part because for casting I need a constant wall thickness, if possible. So mechanical TO does not make much sense there, it is likely deteriorating the casting (for the cast part I will likely ADD material to improve casting). So, naturally, I thought of optimizing runners for casting in the first place, because this makes much more sense for casting, from my point of view. And this is done with CFD TO, and CA cannot do that,

Mario.

mf

jacob
I never asked that.

jacob

well, you asked if it is somehow cfd related.

Are you doing a casting simulation using CA or other software?

mf

jacob
I tried simulation of distortion of cast parts after ejection from the mold with CA. I never succeeded, because todays cast parts can have 30000-50000 faces. To get it right, I'd have to simulate contact of 4*30000-50000 faces (4 times because both sides are in contact with the mold). The meshers in SM cannot mesh such parts, also because they very often have errors in the .stp (with which CAD software can cope with). So, that was that, maybe it's possible in 10 years with 1TB of RAM or even more.
I do not work in this field any more, there is dedicated software to do anything in casting (flow, distortion, etc, these packages can cost up to 200000€ per year).
I can think of TO of runners for casting even without CFD, but somehow the rules of design of runners must be incorporated into such a TO software (don't do sharp turns, certain progressions of cross sections of the runners, minimize distance and or surface of runner etc..). This could be possible.

Nidish

jacob Wow this really looks awesome!
Do you think this can be achieved completely within a single .comm file ? I feel that can speed up the whole process significantly. This is perhaps what the SIMP.comm file above does, although I never managed to get it working.

jacob

No, for larger cases it is irrelevant. And you could not do postprocessing when optimization is running.
A significant effect would have optimization of python functions that calculate new density.

jacob

Hello,
the repository is on https://github.com/code-jacob/TOPyCA

mf

jacob
Hello!

Yeah that works out of the box. Now I have to figure out how to build own studies.

What is great about it, with an intermediate .stl, a redesign of the optimized result is quite easy:

thanks a lot, it's amazing. This is the first time I encounter a FOSS topo optimization tool, that does it all,

Mario.

jacob

Hi,
basically, you just create your mesh with group called REGION (this is the design space), write your BCs to Stage_0_PreProcess.comm and then add them to MECA_STATIQUE in Stage_1_Iteration.comm. That's all.

Next, the whole SIMP setting can be modified in Topology_Optimization.py.
Convergence_Plot.py is just to plot residuals etc.

mf

jacob
Hello,

I managed to get an own example running. But I can't seem to get the MPI version to work with it. At the end of iteration_1 it does write all the files (like VMIS_1.csv etc.) . It tells me file not found in the beginning of the secodn iteration:

FileNotFoundError: [Errno 2] No such file or directory: './RESULTS/VMIS_1.csv'

I don't know why. Shouldn't it be possible just to change the versions in the .export? I'm sure I'm missing something...

Mario.

jacob

in Stage_1_Iteration.comm

But I don't think there will be much benefit. Usually python is the slower part.

mf

jacob
Oh, thanks.

I changed the comments on these two lines, but I am still getting the same error.
Also in the first Iteration, because of the if time_previous == 0: I am not yet in the else: .....
It's this line in Topology_Optimization_run.py

df_1 = pd.read_csv(inp_1, skiprows=4, sep=r'\s+')

that raises the file not found error at the beginning of iteration 2.

It seems to me, that the IMPR_TABLE commands at the end of ...._Iteration.comm do not do what they're supposed to with the MPI version. I have no idea why, the files never appear.

As soon as I switch to 16.2 everything works again. So there must be something else to change. The benefit can be huge when the model is large.

Mario.

EDIT: It seems with MPI, after doing Stage_0_PreProcess.comm it SKIPS Stage_1_Iteration_1.comm and directly goes to Stage_1_Iteration_2.comm. And because in Stage_1_Iteration_1.comm it should generate the first VMIS_1.csv and the other files, it does not find them.

« Previous Page Next Page »