Automated checking of Finite Element Models

[zeroPF]

Hello everyone,

According to this post --> thread507-321928, I am no the only one suffering from quality control of finite element models

I work in a company (aerospace) where I have to check calculations/results of the team .. this gives me a lot of headache

A couple years ago, I started developing (with friends) a desktop software to overcome this particular task: the software must read the input file (bulk data), execute a checklist, and display result of the check.

Simple to say, but not easy to do .. the work still going .. I am alone in the project now (friends gave up)

What do you think about this solution ? Is it worth to continue ?

EM

 

[L_K]

I am interested in this subject also. Which finite element software are you using?
Can you go into more details what kind of input/output checks your result checking
software is doing?

 

[zeroPF]

Some of our customers use Abaqus, others use Samcef, Ansys and sometimes Nastran. My goal is to be able to analyze any bulk data format and see it as a finite element model: mesh + BCS + loads + properties .. and finally perform consistency checks.

Currently, I am analyzing the Samcef format since my main client is using it.

The software is not ready yet .. but I am targeting checks like:

- dimensions of the FEA model in relation to the actual dimensions of the part
- mass of parts
- Are the properties of the materials correct and assigned to the appropriate model parts?
- quality of the mesh
- are the loads and constraints applied accurate
- tests of free thermal expansion and rigid body movement
- ..etc.

I want to be able to do all the checks in one click!

This is a view of the GUI:

 

[L_K]

I use Ansys Mechanical to model mobile machinery. I have been thinking about something similar to what you are doing.
A script that goes through the input (.dat) and results (.rst) files and performs checks. The script would
involve quite a lot of parsing but should be doable. In Ansys Mechanical there is also API which can be used
to interrogate and modify the model.

 

[zeroPF]

I see what you mean! I feel less alone
How do you address the parsing step ? the result of your script are displayed graphically or text messages ?

 

[12345abc6ttyui67]

A black box checking the other black box? Magic.

 

[zeroPF]

Interesting point RandomTaskkk ! can we let it be an open source project ?
By the way, the aim of this tool is to check for human mistakes when setting up input data .. checks should have parameters editable by the user to meat company rules ..

 

[cliff234]

Validating a computer analysis with another computer analysis sounds like a recipe for disaster - because what I hear when I read that is that some think a young engineer with no experience can prepare two models using different software, and if they get the same result, the answer is correct. (Giving a young engineer with no experience a fast computer and the latest software, and thinking they can be as productive as an engineer with 15 years of experience? NO! But that seems to be the way things are going these days.) In my opinion, cursory manual calculations must be used to validate all models. The purpose of the manual calculations is to verify that the manual calculations are within the "ballpark" of what the computer says. In my opinion, too many engineers (primarily younger engineers just out of school) get hung up with too many decimal points. They need to be taught how to manually validate their software results, and they need to learn how to develop a feel for how structures behave. The problem today is that young engineers are thrown into the fire when they start their first job and are expected to be immediately productive. Forty years ago young engineers started out "on the board" and gradually climbed the learning curve. There's a much steeper learning curve today because there's no "board"! Young engineers today have it much harder than when I started out forty years ago. (I apologize for the rant - but this is a topic that I am passionate about.)

 

[L_K]

I agree that this kind of analysis can't possible detect all the possible errors in the modelling. Many of those that were previously mentioned, for
example rigid body motion, should have been caught earlier by the analyst. Other examples include unrealistic boundary conditions etc.

Do the calculation software have an option to open models in read-only mode without license?

 

[WARose]

I'm not sure I'd trust any software to check other software. How does (for example) that software know if the underlying assumptions about boundary conditions are correct? Things like that involve engineering judgment and experience.

Most FEA packages will internally check for instabilities and so forth (STAAD does a good job of that, and adding weak springs in some cases to address those issues).....but that's about as far as I trust them.

 

[zeroPF]

to cliff234 (I apologize for the rant - but this is a topic that I am passionate about) --> no problem I am glad to see that am not the only one passionate about this subject.
Let me clarify one thing: checking a finite element model is a time-consuming, repetitive, boring and painful task. A senior engineer spending his days extracting fields from finite element models to check if the right property was used, if the units are consistent, if there is no negative pressure .. etc, it's a shame to waste time doing that. My purpose is to provide the experienced engineer (usually responsible for checking young engineers work) with a tool which -for each well known check- will:
1- gather necessary data form model,
2- perform the check if possible, if not, display only necessary data for the controller to make the right decision easily

The checklist should be prepared/selected by the experienced engineer. Doing the checks is a job of robots .. not humans!

to L_K (Do the calculation software have an option to open models in read-only mode without license?) --> no need for any license, since the software will read the ASCII input file, parse it to reverse-engineer the model in order to run a checklist.

Remark: I do not pretend that this approach will catch all kinds of mistakes/errors in a finite element model. What I target is input data errors. I would like to be absolutely sure that the data used by the calculation are exactly the agreed data. The interpretation of these data and the results they give is another matter.

 

[MotorCity]

At some point you need a human set of eyes to look at it, otherwise you are checking the results of one black box with another black box.

(RandomTaskkk - the exact same phrase crossed my mind without even reading your post until after I posted)

 

[271828]

I am very leery of this approach.

The magic in a human checking the results-- the old fashioned way, with his eyeballs and brain -- is that humans have judgment, intuition, pattern matching abilities, etc. that programs do not have.

I wouldn't consider an analysis "reviewed" if it was mostly reviewed by another program.

 

[Agent666]

I think zeroPF is using this as an aid to checking, more than pushing a button and accepting the result like many people seem to be interpreting it as. Correct me if I'm wrong.

If some of the menial aspect of it can be automated and then checked independently by hand (sorting the wheat from the chaff so to speak), then why not. Engineers automate design tasks via scripting all the time, so why is this any different if it achieves the objectives required.

 

[zeroPF]

To Agent666: your are right, and this is why this tool must have a 3D viewer to clearly display the model components concerned by the check..so the human checker can control/validate it.

 

[ThomasH]

@zeroPF

What is the purpose of the software? Do you want to convince yourself that the data is correct, of do you want to convice some 3rd part? Would you accept this approach from somebody you were checking if yiou were a 3rd part?

For many of the projects I check the checking is to some extent based on experience and often some kind of visual inspection. Does the model move or vibrate as expected? Magnitude of accelerations or deformations or stresses? Knowing what the results should be (ballpark) is problably better than most automatic checking procedures.

Based on your different posts my suggestion would be to have a look at what you need to check. Then think about methods regarding how to find possible errors in the data. My first idea would not be how to automate but rather how to be smart. If you really mean that you want to do it in "one-click", won't your software be just another black box?

Thomas

 

[zeroPF]

@ThomasH

Imagine you are an aircraft company and you try to redesign a subcomponent to reduce maximum Von Mises stress (say 1000 MPa) at a critical location. You call your favorite engineering partner and you define the strategy and the assumptions .. then you send him all necessary data to build the appropriate finite element model (geometry, materials, BCs, loads, ..etc.).

After 2 weeks, your partner sends you back a finite element model and a nice PowerPoint presentation where 100 slides describe the assumptions, boundary conditions applied,..etc.; 100 other slides show different post-processing plots (deflections, stresses, ..) and the last slide shows that the critical VM is now 990 MPa.

Well, your are happy with these results since everything seems OK (behavior is as expected according to human checking): you accept the conclusion and you pay 10000$ for the good job. The new design is approved for manufacturing.

Now, what if I tell you that the finite element model was wrong, and the correct VM is 1005 MPa? Why? simply because the person who did the pre-processing forgot to tick a box, or add a zero in a field, ..etc.?

To be absolutely sure that all the setup of your model is correct, you must review all input data on the pre-processing software, and you must do that for all your finite element models, again and again. In my opinion, this task should be automated as much as possible, and the human should not spend time remembering that a check should be done in a particular case, or wasting time looking for data in the model. His role in the checking phase should be to review a standard list of checkpoints and decide in seconds whether the checkpoint is OK or not.

 

[rb1957]

only comment with that is "$10000" ?? ... nothing less than 6 (maybe 7) digits !

but to follow your scenario ... if the supplier made an error, then he's going to have his a$$ sued!
The problem for the end user is it's not the supplier's name that the public see or hear, it's their's ...
only more fuel for the suit !

This is a real problem ... how to ensure that the model reflects the designed structure ?
This is why OEMs tend not to extract stress from FEMs but rather load (based on model properties) and then do elaborate ans extensive calcs to apply these loads to the actual structure (and include effects that FEMs don't, like diagonal tension).

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

 

[TLHS]

Honestly, what you're describing is not at all how I would be using FEA in a structural context. FEA is for confirming edge cases that can't be documented realistically any other way. If I'm going to FEA, I already think I know what works, but proving it through conventional analysis would be onerous.

Sure, checking the numbers going in is important, but it that's not how I want that kind of design checked. I want another experienced engineer looking at it completely from scratch because I'm doing something unusual and all of my assumptions are suspect. If I'm stretching to the point where I'm relying on true FEA, I want a check done using a different method, even if it's just an approximation.

If it's beam based FEA, they're already spitting out errors for the types of things you're talking about, so further automation wouldn't be helpful for me there.

 

[TLHS]

Also, the problems that I solve with FEA aren't really repeatable in a way where defining what the inputs or outputs should look like to the computer is necessarily easy. I feel like I'd be spending as much time setting up the checking procedure as if I just had someone spend ten minutes checking for the obvious issues.