ErdemirA 07:21, 10 September 2010 (EDT) Please provide your comments, questions, concerns in this discussion page for Reporting in FEA

GuessT 14:37, 5 April 2011 (EDT)

I was looking at the Validation section and noticed that it is stated "These procedures include direct validation (experimentation done by model developer), and indirect validation (based on data from literature)." I did not notice this before, but I have always viewed direct validation as directly comparing model variables of interest with experimental, and indirect validation as comparing measurable but not primary variables in the model when the variable of interest cannot be directly measured. Nigg defines these validation methods as direct measurement and indirect measurement validation in "Biomechanics of the Musclo-skeletal System, third ed.".

ErdemirA 17:34, 23 October 2010 (EDT)

I noticed the total vs yearly discrepancy, too. See Media:fea_citations.xls. When I did a pubmed search on July 7, I got the number 11,300 (total number of citations with the keyword "finite element", note the quotes). At the same day, I did individual searches for each year using the [DP] filter. When I summed them up, I got 12,002. By the way, when I do a search now, I got 11,709 with "finite element" (note the quotes) and 12,988 without the quotes.

HalloranJ 16:54, 23 October 2010 (EDT)

Ahmet, in the Motivation section of the manuscript I am a little bit confused by the pubmed results. You said that the search string "finite element" in Pubmed yielded 11,300 "citations" in the text but the graph does not seem to agree with this. Also, do you mean that there were 11,300 citations or were there 11,300 studies or "hits", if you will, through pubmed. For reference, I just performed the same search on pubmed and found 12,988 "results", as it is called on pubmed.

ErdemirA 12:17, 29 September 2010 (EDT) Response to Andrew McCulloch's e-mail (see below) by Ahmet Erdemir at 09/28/2010

I definitely agree with you. Commercial software, i.e. Abaqus, has also started incorporating many of the components you've mentioned (every part of the problem) as user authorable entities, except you need to adhere to their not so common language. I loosely follow developments in Continuity and I am happy to hear your long term goal. In practice, my struggle is the capacity to reproduce a model solved with one solver to be solved in another solver. We choose different analysis platforms based on features implemented, convergence capabilities, etc., therefore making models tightly coupled with the underlying simulation software, therefore the software specific mark-up language. And unfortunately, we don't report the models completely for future users or developers to run the model (even using the same simulation software) or export them to another FEA platform or even use another methodology. This is not only a problem in biomechanics but also in modeling and simulation during the product development cycle. As the software develops and becomes easier to use, the modelers decision making process becomes somehow arbitrary and behind the doors. I think each decision should be reported, hopefully with a justification. With our discussions in IMAG, I am hoping that we'll have a chance to link general, self-contained and distributable abstractions of models, with very specific recommendations that are not easy to share but immediately applicable in practice to fill in the gap.

ErdemirA 12:17, 29 September 2010 (EDT) Excerpt from an e-mail sent to Ahmet Erdemir by Andrew McCulloch at 09/28/2010

... I did want to comment that a goal of our finite element package development, Continuity (http://www.continuity.ucsd.edu) is that every part of the problem that would be considered the "model" including global and material coordinate system transformations, constitutive models, dynamic models, models that generate boundary conditions should be user authorable and dynamically compilable datas. With the advent of automated finite element methods (e.g. the FEniCS project, http://www.fenics.org) this concept can be extended to the governing PDEs themselves. This is also enabling us to to develop databases of models that incorporate a complete mathematical description of the model independent of the code.