How can I gain practical proficiency in FEM for a career in FEA?

[bda23]

Hello Everyone,

I've got a background in engineering, having studied quite a bit of finite elements (mainly for solid mech) at university and am thinking of moving into FEA as a career. However, I've always found that, in spite of studying FE books at university and at home, I don't seem to really gain proficiency/confidence in applying the FEM to actual problems. So many factors come into play, e.g. choosing the right element, time-stepping, scaling, etc etc which are not covered in any books I know. Is this knowledge something that only comes with experience working on problems or does anyone know of any hints (specific books or just in general) that could help me in gaining the knowledge necessary for practical analysis in industry? Many thanks.

 

[hadsed]

I have some experience in finite elements/volume and other numerical stuff, though more applied towards quantum systems, fluids and astrophysics instead of engineering. My experience is also rather limited, but here are my thoughts..

Effective use of numerical methods is largely based on how much you know about a particular method's behavior in a given situation. Some methods are diffusive, some do weird things at the origin which you may or may not be able to ignore for your problem, etc. So practicing problems is a huge part of understanding how it will all work. It's not always a pure logic game like analytic problems. Numerics will always be dirty and unclear-- new methods are being devised all the time to account or offset some particular value, negligible to the physics or artificial to the particular method (think viscosity in particle simulation, or diffusivity in grids).

Sorry the answer isn't more clear cut. You can only really get better by just working on more problems. Of course reading more books always helps, and in any case you might want to grab any book that you can to find problems to work on. You might want to also check out any CFD (computational fluid dynamics) or quantitative finance sites (I know of cfd-online or something, and wilmott which has some great resources if you poke around a bit for them, yo ucan Google these two places).

Good luck.

 

[AlephZero]

To quote from another PF thread,

You can't learn to ride a bike by reading books.

You learn to solve problems using FEM the same way that you learned to solve them with pencil and paper. Just do lots of problems. They don't have to be big problems. making ten small models will often teach you more (not only about the FE package, but also about what you are really trying to do) than making one big one.

Any good FE system will have a library of demo, verification, and/or benchmark models (if it doesn't, assume the people who wrote it never bothered to test it - so why are you using it??). They can be a good learning resource.

If you want to use an element type or solution procedure for the first time, always start by modeling a problem where you know the answer. And never forget the first law of computer modelling: all the output from every model is wrong, unless you can think of a very good reason why it's right.

 

[bda23]

Thanks to both of you, your comments were helpful. I suppose the main thing to take home is to practice problems and gain experience that way. I will also try to get to know the particular algorithms in more detail. Thanks again.