FEA meshing - why not 2d mesh -> 3d mesh ?

[Ziggi]

FEA meshing - why not 2d mesh -> 3d mesh ?

Well,

I have just realized why the people do not use FEA (finite elements analysis) as often as they should do: FEA preprocessing phase is a disaster! Simply speaking - there is no simple, efficient meshing software available on the market. Existing software like Hyper Mesh is just an interface catastrophe. Everything you hate in a computer software interface is existent there!

OK - being serious. The problem seems to start at the beginning. Engineering design applications are NURBS driven. They describe surface in a topology unaware manner. Then the project goes into FEA preprocessor where appropriate mesh topology is being created for further analysis.

Well - while contemporary FEA preprocessors are powerful in terms of handling various mesh types, they are extremely poor modelling tools compared to professional modelling tools like 3DS Max or Maya.

Simply speaking - any 3D artists is able to deliver a model with nice surface mesh topology: definitely this model will be better than result of automatic meshing by FEA preprocessors or even better than human-optimized meshes delivered by inexperienced FEM engineer! The only problem is... how to proceed from surface (2d) mesh to volume (3d) mesh keeping surface topology of the input geometry?

Or rather - the real issue is that I never heard about a FEA preprocessor accepting surface mesh data as input geometry. Why? There are efficient algorithms generating volumetric 3d mesh from 2d surface mesh (volume tessellation) but FEA preprocessrs do not implement them. Why?

Would you be so kind to explain why such a streamline is non-existent in practice?:

1) Model construction in 3DS Max or similar application as surface mesh (triangular, tetra or even hexa is possible). Basic topology optimization is achieved already during modelling phase.

2) Convert surface mesh model into volume mesh model preserving imported surface topology.

3) Import generated 3D mesh into FEA preprocessor and apply materials, loads, constraints.

4) Export to solver.

The reason for the question is simple: an hour of 3D modeler (an artist) work is much cheaper than an hour of FEA engineer while 60% of FEA engineer workload is... meshing !

Would you be so pleased, to comment?

 

[shimmy]

Autodesk has a new software called Photo Scene Editor (project photofly) that does most of the meshing on its own. It uses cloud computing, so it's pretty quick. It's in development, so it's still free. It does a pretty decent job if you give it enough photos to work with.

 

[Mech_Engineer]

I've found that ANSYS v13 makes good meshes pretty painlessly these days. I usually make the mesh out of quad-elements rather than tetrahedrons (a simple setting), and you can set things like face sizing, volume sizing, sizing based on proximity to contact conditions, etc...

In parts that are symmetric or approximately symmetric along an axis (take for example a cylinder), you can use the "sweep" feature which will mesh a face you specify with an element size you specify, and then "extrude" that face mesh along the normal axis based on an element size or number of total divisions. In other words, there are FEA meshing algorithms that are face-driven, you apparently just haven't seen them...

As for why meshing is so different than 3-D modeling, they are two different animals solving different problems. An FEA mesh is doing a lot more than dividing the geometry into nodes, it's alo connecting elements with constraint equations, taking into account nonlinear deformation, etc. From a numerical standpoint, how "pretty" the mesh is is basically irrelevant.

 

[AlephZero]

You don't need to do steps (1) through (3) at all. Just define the loads, constraints, material properties, etc on the geometry not the FE mesh, and let the input generator automatically convert them into the FE input.

As #3 says, FE meshes are not meant to look pretty, they are meant to give answers that are acceptable in both accuracy and cost (time) to run the model. Those two requirements conflct with each other, since the "no-brain" way to get more accurate answers is to make a bigger model. The smart way to start with a coarse mesh (automatically generated from the geometry), run it, and refine it only in the regions where the accuracy is too low.

We have been doing FE analysis like that for about 25 years now, using our own software. Some commercial FE software contains similar technology.

The biggest downside was that some engineers were so used to spending more than 95% of their time generating FE models, they didn't like having switch to spending more than 95% of their time actually doing engineering

 

[afreiden]

Engineering design applications are NURBS driven.

...Would you be so kind to explain why such a streamline is non-existent in practice?:

1) Model construction in 3DS Max or similar application as surface mesh (triangular, tetra or even hexa is possible). Basic topology optimization is achieved already during modelling phase.

2) Convert surface mesh model into volume mesh model preserving imported surface topology.

3) Import generated 3D mesh into FEA preprocessor and apply materials, loads, constraints.

4) Export to solver.

CAD/animation software are NURBS driven, but FEA brick elements use Lagrange polynomials rather than NURBS. Hence, why CAD and FEA are always incompatible for solid elements.

CAD and FEA are really designed for different purposes, as the previous posters mentioned.

The transition from one to the other can be seamless when stick elements and shell elements are used. I haven't seen anything for solid elements though..

You don't need to do steps (1) through (3) at all. Just define the loads, constraints, material properties, etc on the geometry not the FE mesh, and let the input generator automatically convert them into the FE input.

What software are you talking about?I do know that Benson and Basilevs at UCSD (and I'm sure some others) are actively working on a solution to the NURBS/FEA incompatibility issue. They call it "Isogeometric Analysis." Check it out.