Study of element types and mesh density 2

[OnlineFEASolver]

For those interested I have completed a basic study on mesh
density and element types using Calculix. The results are
here:

http://www.onlinefeasolver.com/articles/mesh.php

Let me know your feedback and if I should model any other
cases.

Principal -

http://www.OnlineFEASolver.com

General FEA Consulting Services

 

[xerf]

Is your post an ad ?

 

[OnlineFEASolver]

Ad for what? It is an educational study which I am looking for feedback on more cases to compare or any inaccuracies in my findings.

Principal -

http://www.OnlineFEASolver.com

General FEA Consulting Services

 

[corus]

I don't think a simple cantilever is the best case to use as a test of the robusttness of any code, and I would have never even have considered using linear tets. As Calculix is supposed to be comparable with Abaqus, which is assumed to be thoroughly tested, I'd compare the results from Calculix with an identical job using Abaqus. I've used theoretical cases for non-linear thermal transients which compare well with ANSYS results, theory, and Abaqus. For Abaqus I'd be looking for comparative cases that tested the non-linear material behaviour abilities of the code, and not one which tests mesh density, which has been done countless times before to validate finite element theory.

corus

 

[OnlineFEASolver]

The linear tets perform suprisingly well for the fine mesh
cantilever beam. They are also good if you are just looking
at relative stresses. I use them for topology optmization
and they give good results for their price. I dont see a
problem using them as long as you have several
elements through the thickness for a bending load.

My study is meant more for the new analyst looking to
understand some basic info. on mesh density and element
types. This information exists in most big companies but is
not accessible to everyone else.

Are there any benchmarks for non-linear analysis? I suppose
I could just definte non-linear materials for the
cantilever beam and load it till a hinge point develops.

Principal -

http://www.OnlineFEASolver.com

General FEA Consulting Services

 

[corus]

The Abaqus verification manual has lots of examples. You could also try the NAFEMS benchmark tests but for some you'll have to pay.

corus

 

[rb1957]

i haven't looked at your link (so tell me to !) ...
how do your elements fair in single element tests ?

 

[xerf]

If one clicks on the link in your signature (i.e.

http://www.OnlineFEASolver.com)

this is what can be read:


"About Online FEA Solver

We are a revolutionary inexpensive online service to let analysts run their large FE models on our compute servers. We use the open source Calculix FE code to solve your input decks which need to be in ABAQUS input deck format. "


Therefore, in my opinion, each of your posts is an advertisement.

 

[GregLocock]

Links to commercial websites are allowed in signatures.

Cheers

Greg Locock

Please see FAQ731-376 for tips on how to make the best use of Eng-Tips.

 

[corus]

rb1957, I don't think a single element test would be appropriate for a bending problem so any comparison would be worthless, just as using linear tet elements was inappropriate and any comparison totally pointless in my view.

I quite welcome any validation of the free Calculix program so if the results are available on a web site that also offers a service, then ..so what

corus

 

[xerf]

corus,
if you are really interested in Calculix, you might find this book interesting as well:

"The Finite Element Method for Three-Dimensional Thermomechanical Applications" by Guido Dhondt

The author, Guido Dhondt, is also one of the authors of Calculix.

This is from the book description:
"Advanced material formulations include anisotropic hyperelasticity, large strain multiplicative viscoplasticity and single crystal viscoplasticity. Finally, the methods described in the book are implemented in the finite element software CalculiX, which is freely available (

http://www.calculix.de;

the GNU General Public License applies). "

I have this book and I can say that the author has assumed the reader has an advanced math and continuum mechanics background....so, some contour plots on a web-site(which even lack a basic legend) without any precise problem statement cannot be considered FE code validation.

 

[OnlineFEASolver]

My study is not a code validation for Calculix. It is meant to be a very basic study for a new analyst on effects of mesh density and element type on results. Sadly many FE analysts do not have the advanced technical background required to understand the physics and math behind FEA. So my study is meant for that audience, and I can say from experience that audience is quite large..

Principal -

http://www.OnlineFEASolver.com

General FEA Consulting Services

 

[corus]

ffs xerf, they give an error for a simple beam bending problem. I'm not sure why you need a problem descricption, it's a beam in bending fcs. Anything more complicated then, yes, give a problem description. Otherwise the results showed the error you're likely to get with that kind of mesh and problem. It's good enough to demonstrate the accuracy without going to the extremes that some people desire each and every time.

corus

 

[sdra2]

I think I am the sort of person that this kind of study would be intended for, being one of the FEA dabblers whose very existence must be an anathema to analysts. Indeed, I applaud the fact that such information is available.

However, I am completely put off from going any further than the page linked because of the following:-

"....All percentages are rounded to the nearest 5 or 10, i.e 16 becomes 15 and 72 is rounded to 70. FEA results can have an inherent error of 1-2% so reporting accuracy to the nearest whole number is typically not realistic."

I get the gist of what is intended, but why compound an error before you start? I may not be qualified to offer advice on FEA matters (or rather, am not), but I am perfectly capable of interpreting all ten digits in the units column, +/- 1% or 2%.

Furthermore, this range is not conservative compared to other posts I have read here. Methinks that FEA is perhaps even more of a black art than most other engineering disciplines.

 

[xerf]

In my opinion any serious validation case should state the problem being solved including:

-geometry description of the problem domain (problem sketch)
-problem dimensionality, 2D,3D
-statement of the material behavior and the material properties
-statement reprading element type and underlying formulation:
- plane strain/stress, axissymetric, 3D, continuum , structural
- element order (linear, quandratic) and integration scheme (full, reduced, etc)
-any other consideration affecting the element performance
- element size
-clear presentation of boundary conditions
-clear presentation of the loads applied

- statement presenting the time incrementation, for non-linear problems.

- possible statement regarding the solver used

- clear presentation of the reference case used for comparison

- statement of the quantities being compared. The analysist should specify if he used any approximation. e.g. extrapolation from integration points to nodes. If some quantities are compared at certain locations, such locations should be clearly stated.

- definition of quantities being used as the accuracy measures, e.g. relative error, absolute error etc. One should present the formulas used to define such error measures since some may take into account the values sign or not.

 

[OnlineFEASolver]

Thanks, I have added more info on the problem description
here:

http://www.onlinefeasolver.com/articles/problem.php

The purpose of rounding to nearest 5/10 is so the new
analyst should not think a design with 92 MPa stress is
better than a design with 94 MPa stress. For all practical
purposes the numbers are the same in FEA world. The analyst
should treat both results as the same and NOT
make a design decision solely on a relative difference of
2%.

Principal -

http://www.OnlineFEASolver.com

General FEA Consulting Services

 

[corus]

For me there's too much detail, but that's just my opinion. I would add though it was a little difficult to actually find the results from the link without going back and clicking on other links. Maybe a pdf you could download would be of more use and perhaps your general conclusions for 'new-bies' would be of use rather than just quoting the errors. Some people, like myself, see the pictures and then want to jump to the detailed conclusions after seeing the work.
Whether it's advertising or not I still think it's useful to see such verification for this program as I've had people contact me on it's use. More verification would see more acceptance of this program and even a special forum here for queries on its use?

corus