Let me give you a simple example of a true story I have run across.  An "analyst" was doing a "simple" problem of solving the deflection of thin sheet fixed in all DOF on its edges with a pressure load applied.  Seems very simple right ?  

Why then were his results 5X, yes 500% off ?  The answer is because he was lacking the theory and knowledge.  More specificially, he did not understand the concept of stress stiffening in membranes.  

So you see, even what appears to be a simple problem to many may require extensive knowledge.  In my opinion, a Masters degree is a minimum.  This is just my opinion, but think about the above story and try explaining why you got 500% error in your simple problem.

FeaGuru has written : 'a "simple" problem of solving the deflection of thin sheet fixed in all DOF on its edges with a pressure load applied.  Seems very simple right ? '
I think the simplicity here depends on the magnitude of the load and hence the resulted deflection. It is well known that in plane stressed/tensioned elements (not only membranes, but also beams...or in general thin elements!) are stiffener, that is referred as stress stiffening. So a non linear FEA must be undertaken and the option stress stiffening should be also turned on, if the deflection is exactly very larger then expected. But if a "BSc"-Engineer has a long term experiences, then he must be aware about this situation, at least he has read the software manuals (if any), that explains it. Any comments?
I only find it good, if an analyst that does a non linear  or dynamic FEA-problem should have a Masters degree. But a "BSc" one with equivalent experience is also capable for doing it.

cheers

FEAGURU, think about those programs in use out there which don't involve error estimates or that no one is checking!! Yikes!

My point is that how can someone seperate the straightforward problem from the one that isnt unless they have the background in such nature ?  If your satisfied with results that are very accurate sometimes, but garbage other times great...do you know which are wich though ?

The reality is that it happens all the time, I see it all the time and am continually disgusted by it.  I see it by people with Master's degrees and Phd's, but FAR more often from the ones with just a bachelors.  There is no hard and fast rule, but I know which side of the bell curve I'd rather be on is all.  

P.S. Im sure Ive made my share of mistakes as well, you can only hope to minimize them.  

I understand your point well and only meant to emphasize it.  For me there are too many general programs out there that cater to those with less than necessary understanding of FEA yet market the FEA as a tool to be used by anyone who merely picks up the user manual...sorry, I don't agree with that approach.  Nor do I think that FEA is for everyone to work with.

Ok, Perhaps I was not specific enough in my original question. I was referring to FEA theory rather than engineering theory or knowledge in general. I admit that there is not always an easy distintion to be made.

FEAGuru's example is one where it is general engineering knowledge that was lacking rather then FEA theory inparticular. Presumably the analysist would have made the same mistake if using analytical solutions! Also I included the proviso that the analyist could read the manuals and was careful in checking the results.

I do think that properly qualified and trained engineers should be FEA analysists.

TERRY

I think, what Terry suggested, that there isn't a clear distinction between engineering and theoretical / mathematical backround. May be we should refer to the distiction made in http://www.dermotmonaghan.com that engineering = FEA and theory = FEM. That means in order to be an analyst, one should only know about engineering "face" of Finite Element, this is what normally described in many manuals of FEA-Tools concerning limitations of each elements and used solution methodes or the possibility of errors. But in order to be a FEM-Developer or at least PhD-Student in this discipline, then one should have the strong mathematical/theoretical background behind the FEM, such as Raleygh-Ritz or Galerkin methode, shape function, and so on.
From this reason I must say also, that properly qualified and trained engineers should be FEA analysists.

cheers

Well, zuardy, that is exactly what I mean - so why could I not say that in the first place!

TERRY

All Right, All Right..........Let's put the rubber on the road.......

I've got a BS and MS in CE (Structures Specialized in both)and have ~15 years doing FEA in the Aero and civil fields.  The first couple of years I did a bang up job building models and pretty pictures but couldn't get answers that could be validated by a hand calculation.  All the manual reading in the world didn't answer the questions I needed answered.  So, I took a finite element class to learnabout matrix analysis and all that good stuff.  The prof I had was a tenured prof who was a great theoretician but no applicator.  I then took this knowledge to my peers who had gone the same route and then we all began studying what the heck we were doing until we learned that every problem is different and that the greatest advantage is having a great intuition of physical component behavior.  After about the 7th year of this exercise I began to feel "confident" in my anlayses without having someone check my assumptions and procedure.  Now, 15 years later, I beg to have my work checked to make darn sure I haven't overlooked some truly devastating number or input.

What's my point???  It takes 5 to 10 years of very intense  grinding study and application to collect all the tools, skills, intuition, and experience to be an FE analyst.

Want a good test of my hypothesis???......take a job as an independant contract finite element analyst, if you can make it through your customer's expectations and produce a marketable product to their liking AND have them ask you to do more.......you have what it takes.  Try doing that by reading a few manuals..........

ISEAG
Well said - it is true that the best theoretician in the world may not make the best analyst. I agree you do need to be a competent engineer to be a good analyst and this is much more important than knowing lots of FEA theory. However the theory can be helpful but it is more useful than a necessity.

TERRY

"...every problem is different and the greatest advantage is having a great intuition for physical component behavior."

Set that in bold type and put it at the top of the page! The rest you can get with time and a Mastercard, but if this priceless capacity is not in your genes to build upon, you need to find that out as quickly as possible and then find something else to do. If you have a mind that can take a complex structure, intuitively break it into a set of  FBDs, then predict the resultant behaviors, you will probably do fine with FEA. If you never learn what the results are going to be before you get them, then FEA had better be someone else's tool.

Well stated Darrell!

I strongly agree with Darrell et al about physical intuition.

Finite element models are often so complicated that you can't hold all the information about the model in your head at one time. A small inconsistency in one of the 100,000 elements can cause the whole thing to produce funny results (or worse, produce funny results intermittently). My experience is that the majority of modeling errors come from stupid mistakes in the construction of the model, and not lack of understanding of theory.

Whether the bug in your model is that you assigned a concrete element steel material properties, or something to do with convergence, physical intuition is invariably what alerts you to it. Maybe you could argue that further academic study enhances your understanding of of the physical world, but most of the theory we are talking about is just high calibre math. Math and physical intuition are totally different.

I agree with the last few comments. But this applies to any design aid, not just FEM software.

You do not necessarily need to know the theory well enough to write the program in the first place. You DO need to understand the theory well enough to allow you to model the item and interpret the results.

If a designer does not know what results to expect from a computer program (within 5-10%), then he should not be using it.

I had a case about 10 years ago where a soils engineering company did an analysis on a raft slab for project we were doing and supplied us with the results on which to base our design. The person doing the analysis was a well respected Phd in the field. We looked at the results and they looked logical. We then asked for some changes to the shape (architect again) so he reanalysed and gave us a new set of results. A junior engineer in out office then designed the reinforcement for the raft. He complained to me that he could not make the design work in one area, it was having ductility problems and he simply could not add enough reinforcement to make it work. I looked at the 3D moment plot at the point where he had the problem and asked how he could develop a moment at that point, let alone 6000KNm/m. It was perpendicular to a free edge. We questioned the Phd and he fixed the bug in his program for rectangular elements (in the first run he had used square elements and have changed to rectangular to get a quicker solution) and a week later gave us a new output which showed a moment at the same point of 0. He had been using that software for several years and never noticed a problem.

The Phd should have picked the problem. The moment plot "LOOKED WRONG". That is the understanding of FEM that a design engineer needs. An understanding of structures and how they act and what the results should look like. With FEM, there are too many numbers to look at every one and check its vaule and logic. But the designer needs to know what the pattern of results should look like, what the order of the numbers should be.

Computer software is not meant to replace engineering knowlege, understanding and ability, it is meant to do more indepth calculations to allow an engineer to make better decisions about a design situtaion. Too many "engineers" use design software because it obviates the need for them to learn how to design properly and understand what they are doing. They are wrong. They are not engineers, they are technicians.

Many software developers push the ability of their software to do the job of an engineer. Noone should buy or use their software. Software is a tool to be used by knowlegeable people who understand the field. A good example of this is the number of designers who have absolutely no understanding of prestressed concrete design who are putting out designs using Black Box software. They do not understand what they are doing or what the software is doing and the limitations of the design methodology they are using. They are dangerous and so is the software.

I think the last discussions "converge" to the conclution that advanced FEM-Software can not supersede the FEA-Engineer's knowhow. To emphasize this, let's refer to the well known error posibilities in using FEM-Software :
1. Idealization and/or modeling error (physic)
2. Discretization error (nature)
3. numerical error (mathematic)
The FEM-Software actually can only capture the 3rd. error, since this is completely done in the FEM-solver. But due to the influences of the 1st. and 2nd. error (if any), the complexity of the error rises and therefore the FEM-Software can only give a "User Message" to the engineer (if it can!), so the engineer should undertake himself the selective changes to the FEM-Model. E.g.
 - a not properly constrained DOF (1st.error) would call User-Message : singularity for this DOF or
 - too fine meshed dynamic structure would create a numerical error type of User-Message, due to the complexity of algorithm used in dynamic solver.
And yet many examples.
In general the 1st. error is under the responsibility of the engineer and fortunately (^-^) can not be fully superseded until now by the FEM-Software (Preprocessor), since it is a blind tool, so it can't "see" the real problem and can't make the decision concerning the element type and the soulution type. This is a huge challenge of FEM-Developer to create an algorithm that enable the preprocessor to analyze the shape of given structure and then mesh it automatically with the suited elements. At least they can implemented an expert system in their FEM-Software to support the engineer during preprocessing.
The most recent developed element free Galerkin methode or sometimes called meshfree or meshless could make this kind of engineer job obsolete, but fortunately it can't solve  thin or slender sturcure problems well, so it remains the engineers job region, yeah .
I hope this will make clear where the FEA-engineers stand and where the FEM-software stand. We all do not want to loose the job in the near future at least

cheers

My two-cents on this. Nowhere did anybody mention what level of sophistication we are talking about.  Great comments, but I would counter Terry's original posting by stating that at a certain level, the FEA theory IS the engineering theory.
When one starts dealing with the very high-level problems, often the behavior of the structure cannot be understood by the most knowledgeable "classical mechanist", because the frontier is solely in the FEA algorithms employed.  
A good code will state these distinctions within their theory manual, but not at a level which can be easily discerned by anybody short of an advanced applied mechanician.
I agree that for more simple problems, a good engineer can run good FEA without knowing the nuances of FEA solvers. However, as soon as somebody starts stumbling into singularities, zero pivots, negative eigenvalues, wavespeed problems, and convergence difficulties (to name a few), one will have a hard time interpreting what is breaking down without the theoretical basis of FEA.
Thanks for the interesting posting!
Brad

Thanks all,
In my original post I was trying to play devils advocate, but the discussions have helped me to make some sense of the original question. I am now certain that:

1) A good analyst is first and formost a good engineer, this should include good physical intuition for the problem. In particular I think I would sum-up the requirement in that "if an engineer cannot make back-of-the-envolope calculations of the problem they should not be using FEA/FEM to solve the problem".

2) Depending on the complexity of the analysis, a detailed knowledge of FEA theory is not required to be a competent analysist.

3) Any good engineer that is using FEA will familirise himself with the appropriate theory. See point 1).

So perhaps this has been a bit of an accedemic discussion. When hiring your next engineering analysist look for someone with a passion for engineering and an inqusitive mind - perhaps the rest will follow!

Brad, thought we had lost you - welcome back.

TERRY

Terry,
Great summary.

I'm still alive and kicking.  However, the presence of a 13-month old running me ragged (and warming outside weather and "honey-dos" that come with it) are starting to limit my surfing time.


Brad

Tubyguru,

If you get hired buy a small, large or any company straight from school with a BS in engineering, you should not even be designing simply members without an experienced designer reviewing your work. You are an apprentice designer and need several years of supervision before you are a real engineer.

How can you tell if you have not made a modelling mistake if you have no experience as to what the answers should look like?

Do you know how the results of the FE analysis should be applied in design?

In structural engineering, many designers using FE analysis for slab systems still think they will get reduced design moments from an FE analysis compared to a 2D frame analysis. The only thing the FE analysis does is give a transverse distributiion of the moments. The total moment in a panel is the same for both types of analysis. They are simply reading the numbers wrongly and do not realize it and consequently underdesign.

A designer must understand the basics of FE and have the experience to interpret and apply the results, or have direction, advice and supervision from an experienced designer. That is how yoou become an experienced designer. Not by trying things unsupervised until you eventually produce a building that does not fall down when it is built. Clients hate this approach for some reason. You learn from an experienced supervisor who fixes your mistakes before you build them.

Rapt,

I think that most people do actually realise that the moments from a finite element analysis are not "magically lower moments".
As an example, British codes for flat slabs use an empirical method which gives far higher moments than a finite element analysis. This is why some engineers refer to finite element analysis as giving a lower moment.

I would also like to say that the refernces to 10-15 years of experience required for finite element analysis is complete nonsense. Either these people are still typing in code or they do not realise that not all finite element analysis has to be complicated. A flat slab design using finite element analysis is not terribly taxing on the brain is it?

youwhat,

You might, many do not. In supporting my software I have this discussion with many people and you would be amazed how many think that FE will result in less moment than 2D for a regular column layout. I am not comparing to approximate methods such as the BS8110 2-way slab moments, I am talking about properly analysed solutions. The code figures are conservative because they have to allow for an envelope of solutions to allow for span length variations, end connectivity and a lot of other factors. They were actually developed as an envelope of yield line solutions for the variations allowed. They are not elastic moments and cannot even be used for crack control calculations. The people I am hearing are comparing to Frame Analysis of the actual member and its loads.

I have not said that it is not easy to design from a FE output. I am describing 2 areas where designers need experience or supervision in doing this.
1   Getting the model right and the correct analysis results. Experience is needed to know that the results are in the right order/ball park and that modelling errors have not caused erroneous results. An example of this from many years ago with a Frame Analysis is a designer analysing a relatively simple roof frame with sloping members (hipped roof) supported on brick walls. Unfortunately, he put the end supports in as fixed against horizontal movement (the default). Members were very small. On checking just before construction because we thought his members were very light, we found the problem in his analysis. Member size nearly doubled.

2   Looking at how many designers are using results from FE analysis to do concrete design, there are several areas where they are getting it wrong now because of a lack of understanding of both the FE results and design in general. The number of designers who take the FE deflection and multiply it by 2 or 3 and call it a total long term deflection is amazing. They do not understand that cracking has not been considered.
The number of designers who do not realize or ignor the transverse distribution of the moments actross a panel is just as amazing. Deciding that a slab panel is uncracked based on the average stress over the whole width is not logical but it is done continuously in USA design practice.

In my previous posts, I have said that experience is required for ALL design. Not just FE. Otherwise supervision is needed.

The best way to learn about any FEA code is to create your own custom tutorial problems. Make up some simple problems with know solutions that you can work out for yourself using a pencil and paper, then create a few FEA model idealizations. The bottom line is "if you can't solve a problem with a pencil & paper, you will never be able to solve problems with a computer". As a 2nd approach look around on other FEA web sites and download their tutorial problems. Try to duplicate the results using your code. See this site for starters.

David R. Dearth, P.E.
Applied Analysis & Technology
E-mail "AppliedAT@aol.com"

Rapt this is for you

We've built a large number of flat slab structures based on linear FE model output using a reduced E value based on a cracking moment type relationship. This is in an attempt to determine a realistic long term deflection (& creep) behaviour.

As a moment plot will tell you, not all the domain will crack therefore we would need to have a variable E value in line with moment field across the elements. We need to go iterative here. Compute moment field: any cracking? if yes reduce E for those particular elements, then solve for the deflection field. Of course, we can't do this with our package (ROBOT). Currently, we are applying a reduced E, typically 0.28 of short term E, to the whole domain. Conservative (aren't struct engineers always)?

Surely the Yankee design practice and ours (limeys) is the best compromise between accuracy and ease of solution. After all, we are not working for over-funded F1 design teams here!

Any comments greatly appreciated.
Rufus182
talktorufus@hotmail.com