Abuse of the Finite Element Method

[Trying2Learn]

TL;DR Summary

How has it been abused by people who did not understand it?

In general, one could say the Finite Element Method is merely an interpolation method that could be used to solve field equations. Despite that, this question focuses exclusively on the FE Method and its use in Mechanical Engineering.

-------------------

I have noticed that some schools now teach how to USE commercial FE codes (this question is NOT to disparage such codes -- they are stunning). However, sometime I wonder that by not teaching how to write an FE code, that we can invite trouble.

For example: boundary conditions, material properties, inverted elements etc. must be properly considered.

Can anyone provide examples on engineering failures that could have been avoided, had the engineer "really" (and I admit that "really" is undefined, but I hope you get my drift)... really understood the FE method works in mechanical engineering?

Or, to put it another way... Had the engineer known how to at least write an FE code, could an engineering failure (or even an analysis using commercial codes) been avoided?

 

[dRic2]

I'm not much into modeling, but I took a grad course in "numerical methods" last year in which an introduction FE method was provided. It was offered by the math department to engineering only student and lots of engineering students are not familiar with topics such complex analysis, functionals ecc... so the course was made easier wrt the one taken by mathematics students. Nevertheless lots of convergence and stability problems were studied and rigorously analyzed. I don't remember much in details, but the course was a very good start. Plus, if you're willing to work on a modeling project you are likely to take more advanced courses and you'll get better. Thus from my experience I think that whoever ends up doing the job, surely knows what he's doing.

 

[Trying2Learn]

Thus from my experience I think that whoever ends up doing the job, surely knows what he's doing.

No, sorry. My fault. I was not clear.

I WANT to know how it has been abused by those who ONLY know how to use the commercial software. I want to know what failure or difficulties are encountered by hone who does NOT know what happens INSIDE the code.

 

[anorlunda]

I agree with @dRic2 . The additional skills needed to write the software are not engineering, they are math. Numerical methods is the course that best covers the topics.

So it makes the question partially just semantics. Should numerical methods be incorporated in ME courses, or should ME students take a numerical methods course also?

I am the author of a widely used electrical load flow & stability program. The EE content in network solutions is merely Ohm's Law. I like to joke that I built an entire career out of Ohm's Law. But the math of handling large sparse matrices and encouraging iterations to converge, and the math of differential equations, not to mention lots of software engineering, is a lot of work. Those math and SW skills are also irrelevant to most EEs who would use the tools rather than write them. It is like saying that you should not post on PF unless you know how to write the forum software.

 

[Dr.D]

Thus from my experience I think that whoever ends up doing the job, surely knows what he's doing.

Oh, how I wish I could say that!

I'd say that at least 30% of the FEA solutions I saw working in industry and as a consultant had the wrong boundary conditions. Getting the boundary conditions correct is often quite challenging, but it is critical.

In the typical structural problem, it is essential to tie the structure down to avoid a singular stiffness matrix. Too many people think that this can be done this way or that, pretty much anything that holds it in place. In actual fact, to have a valid solution, the boundary conditions and their implementation is absolutely vital. This is for static problems and also for vibration problems in particular.

I routinely read published papers in my own specialty (IC engine torsional vibrations) where static FEA solutions are employed for the connecting rod, completely omitting the acceleration of the con rod CM. I'm afraid I'd have to go so far as to say, "whoever ends up doing the job probably has not got a clue!"

 

[FEAnalyst]

I agree that nowadays universities as well as other institutions/companies put emphasis mostly on the use of FEA software interface. What's more, this software is becoming easier to use. In my opinion knowledge of at least basics of FEM is necessary when performing more advanced analyses (for example one should know the differences between various element types) but there's no need to understand all underlying mathematical concepts of this method and be able to write your own codes. I know several people who are very good analysts but don't go any deep into math. Instead it is far more important to have extensive knowledge about the branch of physics that one solves with FEA (e.g. solid mechanics, fluid dynamics, heat transfer, electromagnetism and so on). Lack of this knowledge may indeed result in terrible mistakes since without it the analysis may be incorrectly prepared and results can't be interpreted properly.

 

[jrmichler]

I used Abaqus for a series of sheet metal forming problems. I selected the element by using the same element as several published papers on sheet metal forming. When my simulation results differed from my experimental results by 30%, I discussed with a faculty member who got his PhD in sheet metal forming using Abaqus. He claimed the discrepancy was all experimental error. The 30% error was several times larger than the calculated experimental error.

I had tested the element by comparing FEA to the tensile tests of that sheet metal, and also by a series of FEA tests involving friction. The software was completely correct in all those tests. I then did an FEA test involving plastic bending, and the FEA had large error. It turned out that the Abaqus element used by the majority of people in the sheet metal forming field had an hourglassing problem that only showed up in the type of bending typical of sheet metal forming.

I completely agree with the above posts discussing the importance of knowing the physical systems, the physics of those systems, and how to properly apply boundary conditions to properly model those systems. I have seen a number of bad FEA results from people with a strong background in the mathematics of FEA, but a weak background in the physics of the systems under analysis.

I have never studied the theory behind FEA. I have worked many test cases, both linear and nonlinear. That's how I found the Abaqus element formulation problem, and also a number of limitations in SolidWorks Simulation.

Hint: In SolidWorks Simulation, do not use the bearing element unless they fixed it in the last year or two.

 

[dRic2]

Oh, how I wish I could say that!

I'd say that at least 30% of the FEA solutions I saw working in industry and as a consultant had the wrong boundary conditions.

I was not expecting this. So I guess engineers do not usually take numerical methods courses... I mean, I took one out of curiosity but I thought that people who want to end up in the field would take several more.*This is one of time where I'm not very sure about my English, is it understandable?

 

[Trying2Learn]

Thank you, everyone. This helps a lot.

 

[essenmein]

Ive encountered the "simulation expert" whos methodology seemed to be to model everything in full detail and wait days for a result that he didn't know how to interpret, if that didn't work, call ansys for a new module that would, and more RAM. This was a CFD problem and I was perplexed about why he would bother meshing fine internal detail of an assembly when none of that assembly encountered the working fluid (air), and a simple dumb box would have sufficed.