FEMs are way,way TOO BIG and Create More Problems Than They Solve and Waste More Time Than They Save 2

[SpaceIsBig]

In recent years the 'capabilities of pre-post processing programs have generated an attitude that if the mesh can be made 10 times smaller then it should be done, regardless of if it is necessary or not !
I blame this on the move away from creating more than one model (each for it'0s specific appropriate task) to a 'single all-singing all-dancing model', simply on the basis that creating and analysing one model takes half the time) cost (and hence half the of doing 2, or 3 or even more.
What a load of nonsense.
We spend I reckon about minimum 75% of our time modelling stuff and maximum 25% actually analysing the results and assessing and doing real engineering, whereas it should be the other way around !!!
Computers may be quicker but not that quick for some of the monstrous creations I've seen hit the CPU.
20 years ago it started in earnest - when I saw an electronic box with half a million nodes being analysed and the whole department had to crash their jobs so tha tone job wouldn't crash over the WE.
Another place I worked had a 'record' of around 7 days for the longest NASTRAN analysis job ever run in the dept. - that was a combination of model size and the fact that 1 non-linear loadcase was needed and about 100 linear - so the analyst decided to run ALL as non-linear to do it all in one job !

So, back to the statement of the post title and ..... DISCUSS

 

[GregLocock]

Hmm, I remember building and crashing 40 vehicles for one program. These days we crash one or two. Perhaps your models aren't very good?

Cheers

Greg Locock


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

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

 

[SpaceIsBig]

Which are the programs for which these days you only crash one or two ? Let me know so I can avoid buying one :O)
My own models are always perfect , and most importantly, not too big, not too small.
The problem generally comes if someone delivers a model which is enormous which you have to integrate into yours (yes, most people don't even read the specs specifying max size requirements).
The problem is often stress people who create an unnecessarily enormous model which then is passed over and has to be used for dynamics too (the dreaded 'one model concept').
Most young engineers have absolutely no concept of a 'reasonable sized model'.
I think they learn about it but then for some reason find it difficult to apply in practice, usually becasue of time restrictions imposed on them. That quickly leads to uncontrolled 'hogging' of computer resources.
Proper education in the matter is one thing, but checking that it's been done is another and it's usually not done unless the person receiving the model 'screams blue murder'.

I'd be interested to read anecdotes from others of cases they've come across.

 

[GregLocock]

I don't know why your organisation doesn't use different element geometries for different purposes. can't help you there. We run 3 levels of mesh density, for durability, crash and NVH respectively.

You may choose not to buy 5* NCAP crash vehicles (remember that is tested by an outside organisation). Your choice.

Cheers

Greg Locock


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

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

 

[rb1957]

I agree, to some extent, that we use FEA like a sledge to open a walnut.

It is a very powerful tool, just look at the different applications where it is being applied. And the software is also very powerful. And we are often "lazy", and will default to a fine mesh because it's pretty easy to do, and less questions will be asked.

A week to run a model isn't the record, well maybe it is for NASTRAN. Back in the 80s we had to pioneer the equivalent of super-elements and running the fuselage look over a week, recording the different runs on paper, recompling, then finding that one step failed and who had to start all over again. These days the same fuselage model runs in a couple of minutes !

I think a 1 hour runtime is extreme !

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

 

[swertel]

I agree with Greg. The first models we setup are big, bulky, and slow. Then we make and test one prototype, tweak the simulation model, and then have a fast and efficient means to test a variety of design variables without the need to fabricate any more prototypes.

Even if the FEMs stay large and obnoxious, spending 7 days on an HPC cranking out matrices is still faster and cheaper than fabricating prototypes, performing a test, and reducing the data.

Maybe your models aren't bad; maybe you're just comparing the wrong metrics. (And by "you" I don't mean you personally, since this is purely a theoretical discussion per your request.)

--Scott
www.aerornd.com

 

[IRstuff]

So, this is a bit of old news, isn't it?

33 yrs ago, when the division was mostly using 286s, our CFO got the division's first brand spanking new 386 machine, because his spreadsheets were taking more than an hour to run. So, he went around boasting about how quick the 386 was. A month later, he was complaining about his spreadsheets taking more than an hour to run; in the meantime, he had added a whole bunch of things that he couldn't run before.

One division nearly tanked our 32-bit microprocessor development because they couldn't wait for the design rule checks. Two weeks later, they found a fatal design error and had to scrap a bunch of material and design a workaround for the material that was already in fab. Our much-vaunted, correct-by-design 32-bit processor wasn't.

TTFN (ta ta for now)
I can do absolutely anything. I'm an expert!

https://www.youtube.com/watch?v=BKorP55Aqvg

faq731-376 forum1529 Entire Forum list

http://www.eng-tips.com/forumlist.cfm

 

[IceBreakerSours]

Just a small perspective from the real world that allows me to pay my bills:

Given the success that computational engineering has enjoyed over the last few decades, business is good for software developers. However, in order for them to continue to grow, they need to grow their target customer base beyond the expert analyst. So, the business of computational engineering is evolving as the stakeholders seem to be pushing for democratization of simulation. That, to me, feels like the experts' workload will tilt more towards guidance, supervision, or more challenging (nonlinear/multiscale/multiphysics) analyses. While the last part excites me, I am scared what the supervisory role would entail.

*********************************************************
Are you new to this forum? If so, please read these FAQs:

http://www.eng-tips.com/faqs.cfm?fid=376

http://www.eng-tips.com/faqs.cfm?fid=1083

 

[izax1]

In the last century, we had to tweak and save elements to get answers. The computers wouldn't swallow anything. That was a very good lesson learnt. You couldn't just use as many elements you wanted.
Learning to see where you could expect stress raisers, bolted connections etc, has helped me to see where to use dense meshes and where it is not necessary. We still need to see where the problems are. That is the best way of QA.

 

[SwinnyGG]

Which are the programs for which these days you only crash one or two ? Let me know so I can avoid buying one :O)

If you refuse to buy a car that hasn't had almost all of the design iteration done in the digital space..... You should probably learn to ride a bicyle.

Except that was designed on a computer too.

Roller skates?

Same problem.

Skateboard?

Same problem.

Your only real option for pure transportation not influenced by the digital revolution is walking barefoot. It's 2017.

My own models are always perfect

Oh. Ok.

 

[ThomasH]

Hi

Is our FEM - models today larger (more DOF's) then they were 20 or 30 years ago. Most likely.

Does that automatically make them better? Not necessarily.

One difference I can see is that today I can build a model without any significant concern regarding its size. That does not mean that the model has to be very large but the software and hardware does not limit the size the way it did in earlier years.

A few days ago I built a model to preliminary analyze the vibrations in a large foundation. It required three nodes and four elements (two springs and two masses). The reason that I used FEM and not hand calculation was that it was quicker.

My concern is not primarily the huge models. It is more the question "What software should I use to solve this problem?". FEM is a tool that can easily become a "black box".

Thomas