I am trying to optimize a circular casing structure for reducing the von-mises stress on it. Many examples of such problems have Minimizing compliance as the objective with mass fraction and/or displacements as constraints.
I don't see this appropriate for my problem. Can someone guide me in formulating this problem? I also want to impose some realistic manufacturing constraint for a realistic final design.
I've attached a short study to show how this can be used. It does give a different solution to compliance based optimisation. You don't say what pre-post processor you are using but it is very easy to set up by editting the input. If you have issues contact MSC support.
The document was very informative and helpful. I have tried to apply the same in my model. But the structure being circular, the optimization process removes out almost all of the material in between and creates a discontinued structure, which is physically not possible. Also tried to put symmetry and axial casting direction constraint. Any thoughts on this?
Stress optimisation will always take more design cycles so make sure you have got hard convergence. Are your loads axisymmetric? If this is something like a wheel then the problem definition is mathematically poor as there is no unique solution. For a car wheel you can have 4 or 5 spokes or more. You could think of using a cyclic symmetry manufacturing constraint? You could also add a displacement constraint at the loading location to ensure everything joins together well.
Attached is the deck of a simple model that I am trying to study to develop my understanding.
Your inputs will be much appreciated.
The major issue is the kind of load. Please note that a gravity load is a design dependent load. You will find some publications on topology optimization wrt design dependent loads, e.g.
Article Effect of self-weight on topological optimization of static ...
Article Topology optimization with design-dependent loads
Have you considered trying NX Nastran from Siemens PLM Software? With Version 13.0 (close to final release), it is now called Simcenter Nastran, and can treat all the considerations you have for your problem, including cyclic symmetry (either reflected or repeated), and it is unlikely that your model will separate into pieces. Version 12.0 will also treat all of the above, but it will sometimes be quite slow for larger stress-constrained jobs (probably not for your small job). Version 13.0 can treat large stress constrained problems much faster and use only a fraction of the disk space. Please ignore the fact that the above looks like a shameless plug :) . My only purpose is to help, so I will ask you to freeze some elements (keep them outside the design region), which may help you preserve the continuity of your finite element model when running with MSC Nastran. Frankly, I do not know how that is done with MSC Nastran, maybe Andrew Main can help you with that. Hard to say otherwise why your model becomes disjoint.
Nils is correct here. An obvious solution to this problem is total removal of all structure. You then have no load, no stress, no mass. Job done! I had a colleague a long time ago trying shape optimisation of a beam to maximise the natural frequency. One of his variables was the beam length, which instantly went to 0.0... A lot of the challenge with optimisation is to get a good initial model setup. I think you need to define a better loading scenario.
Cheers
Andy
Andrew, I am sorry that you interpreted my post as commercial. After close to half-a-century in this field, I am really way past such concerns and only wanted to point out that some issues Amique is having have already been resolved in available software, and that theoretically they are not issues. In fact, I resolved the stress issue over thirty years ago in another, and at that time quite successful, commercial topology optimization software called SHAPE (no one knew to call it topology optimization back then), though it was way ahead of its time and went the way of the dodo when the company founder died, some years after I left. I invite you to take a look at some of my papers, but you would need a clean slate and some unpublished details to include that approach today. The examples I show are sufficient to indicate that stress constraints were handled properly. In any case, it should be obvious to anyone that as the elastic modulus of a linearly elastic finite element decreases indefinitely (i.e. as the finite element disappears), the stress can remain constant while the strain increases indefinitely, so your stress bounds are actually useless if you do not pose the problem to account for that. Adding an arbitrary displacement constraint is totally unnecessary, and may easily change the problem to one that is displacement constrained from one that is stress constrained, so it actually defeats the purpose. There is much more to say, but perhaps another time. Best wishes.
Special thanks to each and everyone for sharing their thoughts on the problem.
Erdal Atrek , I think most of the commands on NX and MSC are similar if we look at QRG, but their could be difference in the way they solve the problem in the background (some efficient algorithms).
Nils Wagner, you have rightly pointed out the critical nature of the Gravity load for this problem.
Andrew Main, I'll try to rethink about the formulation.
Thank you all it has been a productive discussion. Feel free to add more of your thoughts.
Amique Ukani ...Good question… Please check these references below.. hope it will be helpful for you…
http://www.mscsoftware.com/assets/3154_NA2004NOVZTOPZLTDAT.pdf
http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.129.1240&rep=rep1&type=pdf
Regards…
Just a warning, that document is over 10 years old. We have developed a lot since then,
- Badges
- Science topic