Automated checking of Finite Element Models

[SWComposites]

zPF, well, it is one thing to check that your FEA hired gun did what you asked for, but who is going to check that the "strategy, assumptions, necessary data to build the appropriate finite element model (geometry, materials, BCs, loads, ..etc.)" is valid relative to reality and isn't rubbish? Who is going to check that the results from the FEA that was perfectly executed per instructions are valid? Who is going to check that the results have converged relative to mesh density? Checking the hired gun's work is the relatively easy part, and why do you need special software for this? why can't you just use your favorite pre/post processor?

 

[zeroPF]

@SWComposites

You are right: checking strategy, assumptions, challenging data relative to reality, convergence issues .. etc. are what makes numerical simulation a passionate job, and I'll love spending may days doing that kind of tasks. But now, the reality is that I am spending 80% of my time doing repetitive boring tasks on pre/post-processors. Pre/post software (in my opinion) are not designed for quality control, they are optimized for speeding up the model setup .. but garbage in -> garbage out !

 

[ThomasH]

zeroPF
If your models are very similair, perhaps an automated checking approach is meaningful. Personally, my models are normally unique and the way I check them varies even if there are some steps that I always do. But any automated approach will only be as good as the idea behind it.

You say that pre/post software is not designed for quality control. I would say that a good pre processor helps the user build a model with control over the geometry, materials etc. Basically, to make it possible for the user to get the model as intended, and that is quality control. Provided, of course, that the user knows what he is doing.

I wish you luck with your project. Personally, I would base it on available pre/post software and make it as an api-program.

Thomas

 

[zeroPF]

Thank you ThomasH

An other important detail (I don't know if it's usual elsewhere) : we often edit the bulk data files on text editors .. so I can't trust pre-processor databases anymore -> so I have to parse ASCII data files to get the finite element model

 

[ThomasH]

zeroPF
It depends a bit on what you edit in the bulkdata. But some things can be edited from the pre-processor. Of course, it depends on the pre-processor.

But if you have something that can be checked with automation why not automate the analysis setup also. Instead of checking for unit consistency, create a template file with, for example, the materials. I have seen people build complex but very repetitive models from a few parameters in Excel and api-programming.

Thomas

 

[zeroPF]

.. interesting approach .. but this philosophy assumes that the checker is the most proficient in pre-processor software/techniques .. which is not the case for me. I prefer to separate people/tools responsible for creating models from people/tools responsible for quality control of these models. In this way, everyone is free to optimize their tools and methods independently.

Oh yes, to get a clear idea, the software I imagine must do the same thing as Solibri Model Checker Link, but in CAE..

 

[ThomasH]

The checker har to know his tools, just as the modeller has to know his. But why must they be different tools? I don't see quality control of models and the creation of models as independant. As a matter of fact, I don't think they should be completely independant. The final checking is just to verify that everything is ok. The quality control should be in the work method. Easier said that done, I know.

If you checking consiste of numerous repeated mouseklicks, why not recorde them an run as a macro. Provided your software has that functionality.

I sometines use Solibri and it is a useful tool. But as far as I know there is no automation in it.

My impression from this short discussion is that you want to create something that meets you specific requirements, checks the parameters you want to check , and all in one click. That will make you feel satisfied. But for a third part I can't see it as more than a black box. Sorry, but that is the impression I get.

Thomas

 

[zeroPF]

In my opinion, the checker must know the physical product and the FE theory. He can't be proficient in all simulation tools, APIs, software (this is the job of the person who will set up the FE model).

This is why I think the checker should have a new tool whose job is : read (only) a finite element files (Nastran, Abaqus, Ansys, Samcef, .. etc.) -> interpret them to display different views where the checker can say (at a glance) if this meets his requirements or not.

I believe that this approach is general (not specific to my particular case) because each checker can create his own checkpoints or edit predefined configurable checkpoints .... I have that dream

 

[TLHS]

You may be falling into the pit of trying to solve the problem of having too many competing standards by... making another standard

 

[zeroPF]

it's a risk worth taking : I guess I need (only) 6 months to get a first operational beta .. the feedback of the first users will be decisive..

 

[ThomasH]

I agree that the checker must know the product and FE theory but I think that knowing a bit about the software is also valuble. The different softwares have different limitations and so on. But I would like to see the same knowledge for the analyst. The analyst should also know what he is analyzing and what to expect.

Earlier in the discussion I got the impression that you define the analysis, how to set it up etc. The somebody else does the actual actul analysis and afterwards you check that you got what you wanted. Correct? If so, who checks the accuracy of the original setup? In the role I have when checking analysis results, I also check the setup. I mostly work with 3rd part checking. But I primarily work with analyzing structures, checking is part-time for me.

I have not heard about "automated checking" like this before. I have been a member of the forum for several years and I have primarily been at the Finite Element Analysis forum. But I can't remember any discusson along these lines. I can only say that I have serious doubts regarding your approach to checking. Somewhere in this I think you need an experienced engineer to check what is happening. Not just to define how to check the model.

Thomas

 

[zeroPF]

@Thomas : who checks the accuracy of the original setup? --> me, my boss, my client, my "n+1", FAA ..

I understand your arguments .. next step would be to show you a demo ..

 

[ThomasH]

@zeroPF
That means that there is a lot more checking of the project than your automated part. Basically, you are trying to automate a task you find boring. And somebody else will check the complete analysis. What you refer to is what we usually do in the pre/post processors.

Good Luck

Thomas

 

[zeroPF]

Exactly : automate the boring tasks, and only the boring.

Thank you for your valuable advice!