All questions to: meshing, materials, boundary conditions and element properties
1. It's normal that "calcultation with FE rules" doesn't work with T16 meshes or I'm commiting any mistake?
2. How bad is the result obtained through the heterodoxy "force/nodes" way?
Thanks in advance.
yes, you're right.
I've just tried to give up a distributed force to a Tet16-surface and it worked.
How did you the have chosen correspondingnodes? Did you use the option for angle-control or did you select them separately? There are only four nodes per element drawn although there are 10 in reality ... Please try it once to select the whole surface rectangular-selection or the angle-control.
The force-distrubution according to FE-rules is particularly important while using Tet16-elements, because corner-nodes have to get a different value than the middle-nodes of every element. The generated error by dividing by the same divisors depends on the fineness of the mesh. With very large elements the error will be more significant than with smaller ones.
I hope I could help a bit.
What I did was to deselect de option "picking with middle nodes" at the options menu, and then selected the nodes with the window tool (holding the shift key and moving the window with the mouse). In fact the nodes to select were those of the inside of a hole. I applied then the force with FE rules and these were normaly saved and processed with a coherent final result.
If the way of unchecking the "picking with middle nodes" option is the correct one, it would be good if this option is not included along with those ones that need the program to be restarted for its application.
Thanks a lot for your help. Regards
Since I started to try Aurora FEA, I'm dealing with "the mesh issue". On one hand, it's well known that T16 meshes produce better FEA results than the T17 ones (the Theory Guide points this emphatically). On the other hand, TETGEN T16 meshes seem to be irregulars and not always easy to get in the desired size (for system limitations). About its irregularity I wonder if it influences the FEA results, and how. To put some light about all this, and establish the limits of my own system, I made some test series with different mesh types (T16 /T17) and sizes, obtained both from STP and STL models.
These tests showed some things about how NETGEN and TETGEN do their job. These observations allowed me then to establish a kind of "procedure" for obtain clean and regular T16 meshes. In a way is to impose the NETGEN pattern to the TETGEN mesher, working with STL files. Attached an image of the test model meshed with this procedure. The model is of 100 MM high and the mesh of E = 452.670. With the same model I got E = 884.125 where I found the limit of my system.
Furthermore to these tests, I made FEAs for every mesh, included those obtained through the optimization procedure. I did this in order to observe the differences between T16 and T17 results, and, specially, to know the FEA behaviour of the optimized T16 meshes compared with those obtained directly from TETGEN. At this point I observed some particularities that my knowledge don’t allow to interpret with accuracy.
I put all this stuff in a kind of "communication", including the description of "the procedure", and now I wonder if the guys of the Aurora Team would be interested in give a glance to this paper and let me know their opinion.
Thank you all very much in advance.
we have already concentrated on how you can get from the nice Netgen meshes quadratic tetrahedra, too. An appropriate solution will be included in our new version AuroraV2. We do this so that we "easyly" add to each linear tetrahedron the midpoints on every edge. The duplicate thus produced must of course be filtered out again afterwards. Sometimes it takes quite a few seconds long.
So we are naturally interested to read abaout your variant. If you like, you can write an email to us: z88aurora(at)uni-bayreuth.de Maybe you can help us to improve our new program.
Give me please a couple of days for put my studies in english and I gladly will send you them by email.
Best regards, and thanks for my new title of "Power-User".
Actually is not something so impressive as a "new meshing technique", but just a way for assure the integrity of models and manage stl files, that save a lot of time. See please my post "More about meshes" from May 8. In those days I sent a report to the Aurora Team, with no news. After this, I sent two reception requests, also without news. It seems to be a matter of servers, but the fact is that I couldn't communicate. Anyway, if you email me to srglpz23 (at) gmail.com, I'll be glad for sending you the information.
I can remember that you wanted to send a mail some time ago. I read it still, unfortunately, not yet. It may has been classified as junk on our server. Or it has been blocked due to some appended data. I'm going to look through the appropriate protocols.
Your idea, to connect the two mesher Tetgen & Netgen sounded very interesting.
So you can stay relaxed. I'll call again tomorrow at the latest, whether it is necessary to send the data again.
And thanks for the lively discussion here. This is the real meaning of our forum.
Thank you very much for your concern. After that initial report, which wasn't received, I continued my observations and added some other precisions. So it would be better if I can send you the final version.
I found your mail in the Junk folder actually. Sorry about the delay.
I've been looking at your thoughts and experiments. - And I'm impressed a lot of hard work you've done to you! Today I would like to summarize some of them, and add my opinion. Thus, other users will benefit from it, too.
1. The most important point was the following:
Do tetrahedrons with middle-nodes calculate better than linear tetrahedrons? How big is the difference? And how should we decide to get reliable results?
You're absolutely right. If one uses the same number of linear tetrahedra and quadratic tetrahedra, the quadratic always lead to the better results. We have even done some measurements. And there was always a measurable difference of up to 10-20%. This difference becomes smaller and even disappears when the number of elements increases. At some point the mesh consists of a million elements, and the result shows almost no more errors. But that's not the sense of FEA. Your recommendation, prefer to use instead Tet17 Tet16 is mine, too. But beware: The FEA calculation-speed is dependent on the number of nodes - not the elements! A single Tet16 is exactly as complex as eight Tet17 to calculate! One can divide into eight quadratic tetrahedron that is linear, with no need to add a new node to. So the winnig of accuracy is expensive. Maybe you can invest the effort in a finer Tet17 network?! I want to say: element types both have their permission be there.
2. Tetgen can generate Tet Tet16 and 17th Netgen Tet17 only. How can the good element of quality for use Netgen Tet16?
The necessity to create a possibility for the transformation, we have also been identified. Your idea of this is very clever. Nevertheless we have not, however, gone this way through netfabb and the netgen-gui. We have decided to easily add new nodes to rely on the midpoints of each edge of a Tet17. Subsequently, possible duplicate nodes are discarded. The mesh-quality of netgen remains exactly as good as for Tet17. But the calculation-quality Tet16 is yet available. This feature will be includet in AuroraV2, wich you can download on friday.
3. Another feature could be interesting for you, too: Local mesh refinement.
If you already have a crude tetrahedral mesh, you can select individual elements or a whole set of elements, which are then divided into eight new elements. Thus, the mesh still remains conform, a transition zone is created that connects to the crude meshed region. The method works for Tet16 and Tet17.
4. The quality of meshes and the quality of FEA-results in general.
There are many rules on how to assess the quality of meshes and elements. But be careful: The quality of a mesh is not dependent on loads and boundary conditions by definition, because they are simply unknown at the time. The quality of the calculation result is very highly dependent on it. To really make reliable statements about the FEA-results can, I would recommend that you count on several meshes and then compare the results. We currently do research on recommendations for deriving sensitivity. But this is a very delicate subject.
I hope my answers are of interest to you. Otherwise, we're always pleased if you participate in the discussion.