FEA Data 1

[hlustosa]

Hello Everyone

I wish to ask you some questions related to how you use the data generated during your finite element analysis. If you cared to answer them, it would be a huge help to me.

1- Most people I've talked so far tell me that visualization is obviously an essential task. However, how often do you guys conduct some sort of analytical query on your data? I mean, do you solve the model and only visualize the results or do you execute some sort of scripting to derivate more data? And if you do so, how do you do it? Do you create a script or program to access the files in which you maintain your results or do you use another kind of tool?

2- A confusing point to me is related to the mesh. Suppose a simple example where you have a PDE and a bidimensional spatial domain. Let's call u the solution of the PDE and assume that it is a scalar field. You divide the domain into triangular elements, so you have points and edges connecting points. The solution of the PDE is approximated by the summation of functions that you do know how calculate, and theses functions are associated to the edges of your mesh. The solution that you want to visualize is the scalar field u. You can't have the value of u on all possible positions of the continuos domain, so you define a granularity or a set of points in which u is going to be solved.

What I want to know is:
- How is the relationship between the solution you want to visualize and the mesh. Do you calculate the solution only on points defined by the mesh, or do you calculate anywhere on your domain (regardless if it is a point on a mesh or on and edge or outside both)?

-The solution in this case is a scalar field. You have a set of points xi and a set of values of u evaluated at every xi. I'd say it is a point cloud with associated data values. However, I've read some papers about data structures to represent FEA data. And they talk about data related to higher order mesh elements, like edges, polygons, polyhedra, etc. I understand that the solution of a PDE would be a (scalar, vector or tensor) field, i. e., data related to points and not to edges or polygons. So, what data is related to higher order mesh elements? Are they inital parameters defined before the solution or even values derivate from the final solution? I'd need an example!

Thanks in advance

 

[GregLocock]

In modal analysis there are tools available that quantitatively assess the agreement between the FEA results and the real world results. They don't, in my experience, help much. So we describe the mode shapes from the animations, record the frequencies and see how the tables compare for FEA vs lab. if you can get the first 6 modes for a bodyshell in the right order with roughly the right frequencies on the first run you are in the top 5% of modellers.



Cheers

Greg Locock


New here? Try reading these, they might help FAQ731-376

http://eng-tips.com/market.cfm?

 

[aerostress82]

I have a feeling that at some point you are also questioning the necessity of FEA.
PDE (Partial differential equations) are perfect for easy geometries such as rectangles and maybe hexagons & etc in a domain.

The beauty of FEA is that you actually have a whole domain of PDE's - each element representing a PDE domain. So, what you would be doing is, depending on your needs, you would model your elements for higher accuracy where necessary. The necessity of the task would call for it as you faced them during projects. For example, you wouldn't want to model a stiffener as a BAR element and model shell elements right nearby (sharing same nodes) in some cases. But again, in some cases, you would be fine with BAR elements connecting to your shells. So, it is really up to your needs at all times.

I've performed a forced frequency response analysis with a moderate level complexity of quad shell to tetrahedral element connections "via dummy shell elements with large bending inertia values and RBE3s" (ignore the quotes in this sentence if it is confusing). What I came to find out was that the accuracy of the "stress" passing through those connections would go down by 20-30% when using higher order elements. So, there is that using higher order elements dilemma. In some cases, you go with higher order, but most probably the first order elements will be enough for your case (simple tria/quad elements with 3/4 nodes).

How the stress is read through these elements really depends on what you are looking for. In some cases, you will just be fine with element stresses in which you will see each element colored differently depending on your its stress, where in some cases you will want to go with "interpolation" points to derive the stresses to see a smooth passage of stresses on the whole structure to observe the structure stresses. In every case, the origin of FEA is the stress calculated at nodes. So, whatever you would be looking at stresses no matter how you derive them depending on your post-processor, you could derive them yourself using PDE.

The only and the most important thing to be careful for FEA is that, you always "always" have to know what kind of modeling affects your results for better or worse depending on your linear/nonlinear/modal/transient/frequency response/implicit/explicit.

Please let me know if I misunderstood your question. I'll write another reply to get it better the next time..

Spaceship!!
Aerospace Engineer, M.Sc. / Aircraft Stress Engineer

 

[rb1957]

my 2c ...
1) there are two main ways to access the results of an FEA ... graphical using software (like FeMap or PATRAN) that are post-processors for displaying the results is a very easy to use format (often called "cartoons" because it's easy to make meaningless pretty pictures) or brute force, going through the output file from the analysis code (like NASTRAN), sometimes using a processor software to combine the results together into something meaningful for the subsequent analysis.

2a) yes, the equations are solved for the points (nodes or grids) that the model is approximated to (ie the real part is replaced by it's mesh ... it can be a coarse mesh or a fine one)

2b) higher order elements (with mid-side nodes) allow for more complex functions along the edge which will typically give a better solution. For example, if you have an edge in you mesh, the two end nodes limit you to assume a linear function; if you add a mid-side node, the three nodes permit a quadratic function. a linear function is usually fine for very simple geometries. but complex geometries require either a fine mesh of higher order elements.

another day in paradise, or is paradise one day closer ?