What's the Real Value of FEA in Engineering Design?

[cmmcnamara]

Hey all,

I have a question that has been bothering my quite a bit lately. Please excuse my relative ignorance of the topic too while I babble

I am an BSME senior at Cal Poly with one year to go. Part of our curriculum is FEA done with FEMAP and I have yet to take the course (next year I will). However currently I am taking a technical elective which doesn't require us to have taken the course but nevertheless incorporates elements from every engineering course offered and meshes it into one giant machine design project. So essentially everyday is a bit of review/crash course on some engineering topic and then project work.

My professor was giving us the review/crash course on FEA and how it works behind the scenes. What I took away from the lesson is that FEA results are generally crap. That in general FEA underestimates the stress quite a bit and getting the best approximations you can takes years of experience to identify and interpret the results. Essentially, and I'm paraphrasing, "FEA results are meaningless without verification of the results with hand-calculations coupled with a grid-independence study." And I'm not one to doubt my professors, especially this one being the certified bad dude he is.

Which brings me to my question: What's the point of FEA?

I realize that computer results allow us to move through things quicker (or sometimes not--it sounds like FEA solutions can take days or weeks to calculate if large enough), but it seems to me just to be an added step to create a silly visual display for the customers/managers since the results are not what the actual designing criteria gets based off of. Is it really just visual way for us to get a point across at a single glance? Or are the results really actually useful? Is my professor's outlook unique on this topic?

I'd really appreciate any others input on understanding this topic and insight. And before you ask--no I haven't addressed the issue with my professor just yet, it's been a busy quarter!

 

[AlephZero]

Essentially, and I'm paraphrasing, "FEA results are meaningless without verification of the results with hand-calculations coupled with a grid-independence study."

Hm... I've been using FEA in industry for several decades, and I've written some FE analysis programs to handle problems that were beyond the capability of any commercial software at the time I wrote them.

IMO either your prof doesn't know what he's talking about, or you got the wrong message.

The right message (and it's an important one!) is' "FEA results are meaningless unless the analyst knows what he/she is doing, and is using FEA software that has been properly designed and properly tested".

I don't have any experience of FEMAP, but I do know that MSC/NASTRAN, one of the industry standard FE vendors, has about 1200 employees world wide, including about 300 PhD-level engineers working on software development and new analysis methods. And their annual sales revenue is of the order of $200m to $300m.

The company I work for pays MSC something like $100,000 a year to use their FE software. We don't spend that sort of money just to get "pretty pictures" that don't mean anything!

 

[SteamKing]

FEA allows one to rapidly model a problem and obtain results which could not otherwise be obtained by hand calculation. In addition, FEA can be used to refine a design to reduce cost or to determine areas which need additional material to reduce stress. This is not to say that FEA completely eliminates the need to build and test prototypes, but it can help to reduce the number of test prototypes which need to be built.

Certainly, your professor is right in one respect: If a piece of software is poorly written and not tested sufficiently, then the results should be viewed with a healthy amount of skepticism. Ditto with the work of an inexperienced or untrained analyst.

Some FEA models are quite large and intricate. Entire airframes and vessel hulls can be modeled nowadays, but most of the time spent on FEA is spent on generating and verifying the model and loading description. Computers are now fast enough that even large models can be run in a few hours at most. With additional tools, a lot of the model development can be automated, relieving the analyst of a lot of the drudgery of compiling the input data.

FEA is pretty much SOP for structural engineering, mechanical engineering, fluid and aero dynamics, etc.

 

[Chestermiller]

I concur wholeheartedly with steamking and alephzero. FEA is indispensable in industry for doing sophisticated structural analysis. Doing modeling using FEA or any other modeling tool is like playing the violin. In the hands of someone who knows what he is doing, it can make beautiful music, but in the hands of a hack, there are going to be lots of squeaks and discord.

Chet

 

[timthereaper]

Couldn't agree more with what's been said.

In my experience, it's been indispensable. The main thing about it is that you have to have a good understanding of machine design principles in order to be able to build the model and interpret the results. If you don't understand what is happening in your simulation (i.e. how/what/where loads are being applied, what/where constraints need to be applied, etc.) enough to model it accurately, then your results won't mean anything. Another thing is that you have to understand the basics of FEA, like what each element type does and where it can be used, or what is generally happening under the hood with the matrices. You can't just sit down with an FEA package without any knowledge and model up something and expect it to deliver good results.

As far as approximations go, it's not too bad. I've run a few simple test cases on structures I can analyze by hand, and comparing the results I'm often surprised how close the approximation is. However, a new branch of numerical analysis called isogeometric analysis (IGA) is supposed to be even more accurate than traditional FEA (mainly due to using better basis functions than just first-order functions). Plus, it can operate directly on CAD model geometry without making the analyst create and mesh an entirely new model just for analysis. I'm interested to see what comes out of that research.

Anyway, that's my $0.02.