Hello everyone !
I found out on Google a interesting article about tetra vs hexa elements, the author compare the results performing linear and non linear analysis, compression, tension, bending, torsion, vibration and also perform real tests. It really worth it reading ! (If you are interested of course).
Please feel free to post your conclusion/opinion and share information !
Yes, very intersting, Rodrigo! But did you also do any study on efficiency. I'm thinking about time and Computer resourses. If you have model where you could get acceptable results with say, 1 million nodes with linear Hexas, how many nodes would you require to get acceptable results with quadratic TET's?
In terms of numbers i have no idea, it varies from model to model, geometry and so on, But to figure out this you have to perform a convergence test.
Gradually you will raise the number of nodes in your model and check the stress level, if the stress diference between the tests are low, you are in the sweet spot, which means you dont have to increase the number of nodes.
Did i answer your question?
Yes, that's true. I'm no fan of Tets, probably for no better reason than when I started usinfg FEM back in the 1980's, using Tets were banned and was not an issue. But I see the point that it is, for many geometries, much easier to mesh. But as long as I can mesh with HEXAs with reasonable effort, I tend to stick to that.
Indeed, the pros for tetra elements is that when you have complex geometry it´s almost impossible to mesh using hexa. In my point of view, banning tetra elements is a mistake, as i said, tetra elements plays an important role. But if you are able to mesh using hexa elements, of course, you should use it.
Regarding the convergence test, i wouldn´t say that, no matter what software you are using (FEMAP,ABAQUS, HYPERMESH, ANSYS...) "convergence test is the basis theory of CAE" (A. Filho, " Elementos Finitos A Base Da Tecnologia CAE ". São Paulo, Érica, 2013.
I don't think there is anyone banning TET anymore (parabolic that is). It was impractical decades ago because of the amount of data to deal with and computing power available, but Model Solution, NX Nastran and others have been using iterative solvers for 15-20 years, so the performance has been just fine. The issue still remains of the size of the result sets. There is no problem getting good results with TET10 as long as they are used properly. Same is true for HEX to some extent, but it is so easy to generate terrible data with a TET mesh...
The issue is rampant laziness or ignorance, facilitated by this so-called "democratization" and the new illusion that 5 minutes on youtube and asking reddit is pretty much enough to "do analysis": as long as you can catch a few buzzwords to use in conversations to sound knowledgeable and know when to run away or pretend enough to get by... I see it everyday and it is sad the amount of garbage being generated...
The basis of FEA is Ritz, Galerkin and Euler approximation of continuum. In other words: there is a lot of high level math involved that also comes with major caveats. Unfortunately, the "democratization" gives the illusion that understanding any of this is not neccessary...
Here's a great video that could illustrate my point:
Shinji Kagawa and his teamates use bricks & TETs when it makes sense to obtain good results