Finally I have found some time to make a small SnS test. I have taken the steel part from my last post and checked it in Straus7. The brick mesh I made in Straus is quiet nice and clean and the SnS resolution is reasonable.
Now the results: The deflections are similar in size, but the shapes are a bit different (the zero displacement spot in Straus is in the middle of the plate, in SnS a the upper edge of the plate). The stress peaks are at same places, but the stresses are very different. Checked loads, checked material, checked restraints, checked model, checked analysis type - all the same between SnS and Straus.
Now I am a little bit confused, which one of the results is closer to reality? What could cause those differences?
Sorry, I thought the images will show in full resolution when uploading. I will post them at the gallery section.
The SnS stress peaks go up to 3100 MPa, the Straus ones up to 2100 MPa - quiet a lot of a difference. The deformation scale is the same for both pictures - you can see the huge difference in the shape of the two parallel plates.
I agree with you that the results are quite concordant. Maybe I am misunderstanding the purpose of SnS as a helping tool at the design phase and am taking the numbers too serious? Anyway, as an engineer I would expect some result differences within max. 10%...
I will give it another try at something different, maybe this example is tricky...
As Matteo already mentioned, qualitatively the results appear quite similar.
There are many reasons why results could be somewhat different. First and foremost, remember that Scan&Solve computes on original geometry, whereas mesh-based FEA computes on *approximation* of this geometry. Secondly, Scan&Solve applies boundary conditions to faces, whereas mesh-based FEA apply boundary conditions at nodes of the mesh. And finally, resolution is counted differently in the two programs.
Having said this, Scan&Solve has been benchmarked on carefully selected problems agains high-end FEA codes, with consistent results. In general, under identical conditions, you should expect very close match in displacement values and locations. Stresses are more difficult to compare, because they can vary rapidly, particularly near stress concentrations (20-50% is not uncommon!) and are very sensitive to numerical computations.
It is hard to say more without getting into nitty-gritty of your particular example. Also, you should run the problem at different resolutions, in both systems, to see if you are getting consistent results.
I was also thinking about this matter and came to the conclusion that the stress peaks are something that is really not relevant for the overall picture of the analyzed part. In most cases those few peaks have to be hidden to see the overall picture of the part (not just a blue painted part with two red dots). This gave me an idea for the "wish list" - to have two sliders at the results bar which would define the range of results we want to see.
Anyway, SnS is really amazing and I fully respect the huge amount of work put into it! I will post some more examples, also with gravity.
Was wondering whether (providing it does not present any confidentiality or IPR issues) you could post the geometry as an IGES or STEP file and I can then run the same problem using ANSYS. We can then do a tri-FE comparison.
One thing I noticed was the stress contours from the Straus7 analysis contain sharp discontinuities which would suggest that some mesh refinement might be required.
No problems with the IGES file, I will post it a bit later with the other data (loads and restraints).
Sure, the Straus models may be not perfect. It has approx. 74.000 bricks and is cleaned up several times to erase all of the microscopic elements. This was the smallest amount of bricks to get "round" holes and fillets without any specific time consuming mesh refinements.
Please note that the idea of this was a quick experiment to see how things work, nothing more and nothing less. I have just posted what I have experienced...
Vadim, should we continue with this issue within this forum, or is it getting out of bounds?
Why? Personally, I think that this is great. But it is really up to you two. If you want to take it offline and come back with results, it is fine too -- but I am enjoying this :-)
Unfortunately I do not seem to be able to download it. I get the following message
You don't have permission to access "http://storage.ning.com/topology/rest/1.0/file/get/2506643318?profile=original" on this server.
Can you save it as a rhino file and upload that instead.
Here are the results using S&S and ANSYS. My displacement contours appear to be slightly different to yours although the maximum amplitude is about the same.
In S&S I have used the lowest resolution setting and in ANSYS I have used about 30K element. The percentage difference between ANSYS and S&S is about 6% (displacement) and 10% (stress) for the current mesh setting. However, the effort involved in setting up within ANSYS is considerably greater (about 500%) compared to setting up within S&S