All questions to: meshing, materials, boundary conditions and element properties
You have done some really good work!
With a minimum of elements the results are good enough in my field of work.
especially if the models gets bigger its a real advantage.
Does the accarancy increase when you try it with more clearer "plate" type of problem", for exampe a thin walled tube of 3 meter.
How did you create the elements!!!
I've been trying to get the submarine shells for some time without any luck..
I'm really interested in the details and files!
Concerning our subject, I didn't find the way to get other displacement output different from 5.68 MM. See you... I tried the following:
a) To refine the mesh, up to 10 MM (the initial was of 35 mm).
b) To make it coarser, up to 70 MM.
c) To apply the load not as a "Line Load" but as an "Uniformly Distributed Load" by dividing the 8000 N among the number of nodes of the line (what is not right).
d) To apply the 8000 N to a single central node.
e) To apply the load as a pressure, having previously created a delimited area of 25 mm wide, and being the pressure 3.2 N/mm2 (8000 N / 2500 mm2) (see image).
And... always... ALWAYS !!!, the result for the deflection has been 5.68 MM ! It seems as this value is nailed somewhere among the stiffness matrices and there's no god capable to modify it! At this point I don't know what to think, if E 24 (22) shells from common CAD files are very consistent, or if they are useless at all.
About your accuracy requirements from FEA, related to the products you deal with, I understand them very well. In my case (chirurgical implants and their associated instruments), being its integrity a critical subject, there are not a too high exigences for the FEA outputs. This is due to two reasons. On one hand nobody certainly knows which are the real loads. Normally there are several possible and simultaneous load arrays, where constraints are illusory (you know... the "floating objects" that we discussed once). On the other hand, according regulations, all designs must to be physically tested for its approval. No withstanding this, AURORA FEA has become for me a fundamental tool, not for providing final guaranties about the structural integrity of a part, but for providing valuable information along the design process. This is mostly to see how stress is distributed under different load hypothesis, and then take design decisions for avoid dangerous concentrations. Finally, the sizes and materials are highly limited due to anatomical an biological reasons.
Concerning solvers, yes, for every calculation I first apply "turbocharged" PARDISO. Normally, if BC's are correct, is just to click "start calculation" and see on the left that so nice window saying "Solverinfo: ran successfully!". (Note: the PC where I run AURORA is not exactly a racing one). Congratulations to the AURORA TEAM!
- TABLA04.jpeg (59.66 KiB) 10358 mal betrachtet
That's a very important fact in designing/creating products. Always check your results by other methods and apply sanity and reason. FEA is just like structural optimization a design tool which is not meant to substitute (the brains of) the designing engineer. Thank you for this comment!selopez hat geschrieben:On the other hand, according regulations, all designs must to be physically tested for its approval. No withstanding this, AURORA FEA has become for me a fundamental tool, not for providing final guaranties about the structural integrity of a part, but for providing valuable information along the design process. This is mostly to see how stress is distributed under different load hypothesis, and then take design decisions for avoid dangerous concentrations.
Thank you very much!selopez hat geschrieben: Concerning solvers, yes, for every calculation I first apply "turbocharged" PARDISO. Normally, if BC's are correct, is just to click "start calculation" and see on the left that so nice window saying "Solverinfo: ran successfully!". (Note: the PC where I run AURORA is not exactly a racing one). Congratulations to the AURORA TEAM!
I got the meshed beam and instructions from selopez and played a little bit with it.
i changed the plate thickness and it gave results which are similair as the results lopez published.
also i tried an other load case, beam fixed to one side and a force of 800N on the other end and the results were similair.
with similair results i mean the hand calculations give bigger deflections but the FEM results are close enough but always with less deflection.
And close enough I mean far better compared with equal amount of thetraeader elements or no1 elements.
My big question is how to build these nice elements?
Hello frans,frans hat geschrieben:My big question is how to build these nice elements?
Could you specify your question? Which elements do you want to build?
I would like to make e24 elements from easy *.step files.
just to make a metal box, tube, plate frame and other simple forms with e24 elements.
maybe somebody has a suggestion?
The method you refer to, of producing E24 shells from .stp files is only at this time a particular 'taster recipe' from our good cook 'selopez' and is a very new concoction . It's being studied by the Aurora team now- see his post 'THE YELLOW SHELL-MARINE' under the 'how to' section.
I can't help you anymore than i have already for, apart from my posts, i have no direct experience with shells. I gave you a link to the youtube FEMAP using plates showing you how it still takes knowledge to get a good result and some of the pitfalls to watch out for in using / joining plate shells. If you raise your eyes maybe just 100mm up and to the right, that crankshaft mesh can be achieved on 9 year old medium spec workstations. Today, 3 or 4 year old workstations can be bought from ebay at 300 Euro's - even quad cores are available on Ebay for that now. Now i'm not saying it would be done in a snap - no! it would take about 3 hours with a dual core m/c i expect. So amazing meshes can be processed with minimal computing power providing the mesh is sensible / computable - and that comes from experience which as i said from an early post is from trial and error.
However, i suspect your main problem is lack of time.
All i hope is Aurora team can help with the experience to get you going, lack of time is something all of us suffer from.
what is the exact status of the shell elements?
i found a easy way to mesh shell elements using gmesh.
based on this methode.
here a picture of my fiddeling around.
not checked any results yet, i'm just so happy i got it imported in aurora.
- (93.99 KiB) 6275-mal heruntergeladen
This is the first 'plate' design i have ever used FEA on. I have done this to prove to myself the practicality of using solid elements in typical sheet / plate designs.
This is just an example of what can be done with a nine year old workstation, single 3.6Mhz Xeon 2Mb cache
2 Gb ram 512 Mb Nvidia Fx quadro 4500.
It took 2 m 30 sec to process using pardiso solver.
The size of the part is 2 m long , square tube 200mm x 200mm wall 2mm with thicker flanges top and bottom.
Additionally, i've added a bracket which is 0.5 mm off the tube sides ( not touching ) with attachment welds fully modelled attaching bracket assy to the main tube.
So , think of the surface area of the plate material - lets say approx 2m x 0.8m with thickness of 2mm - All
meshed in 3d solid elements. I believe this ratio of sheet area to sheet thickness would cover many real world produced articles. Office type metal cupboards are around 1mm and machine tool covers are around 2 to 3 mm thick so this example assembly is in the right ballpark.
This assembly was meshed outside of Aurora in Netgen 4.9.13 and input via an Abaqus file .inp - Mesh checked -'computable' . Pads on the bracket top plate had in total 1000N applied as surface load.
I do see that a simple surface mesh is easier to manipulate on the screen but i hope this example gives you confidence as to what can be done in Aurora with basic computing power.
Interestingly, i see instantly where i can remove material and where i can make improvements although it is not highly stressed with 1000N load.
To model the design, mesh it, import to Aurora, apply LBC's etc and report / post here has taken about 4 hours.
A great design aid!
ps. the m/c you have ( as your post) has 4 cores to my one and 8 x the RAM memory - it's a very capable computer. If you're having troubles, perhaps it's the graphics card trying to manipulate all those tetrahedron - try hiding graphic info with the white cube button or, pan, zoom, spin etc with only wire frame - it does help.
Also, try adjusting the mouse velocity under tools>option>view - for each zoom scale you're using.
Also, i have given up on using the mouse wheel for zooming - it's so slow! Instead, i use the zoom setting tool and change it a few times during a session. (Sphere with arrow surrounding it).
I uploaded the wrong results,
results should be around 0.619mm
fem results are 0.539mm, 0.567mm, 0.568mm as shown in the pictures
i'm very happy with these results, and i posted the wrong results in my enthousiasm...
In my above example the tet meshing of the 2mm thick plate is very coarse and is not to be trusted for accurate results.
In making the mesh much finer for better accuracy my computer gave up - just too many tetrahedron. So, i reduced the overall length of the tube drastically.
The mesh length edge is around 4mm . The wall thickness is 2mm.
Applying the same load as above example similar results are obtained.
From a design point of view, both examples showed correctly where the high stress was concentrated indicating where design alterations should be made.