# FEA vs Traditional Conservative analysis

### Help Support Homebuilt Aircraft & Kit Plane Forum:

#### proppastie

##### Well-Known Member
Log Member
Those of you that do FEA, has there been any comparison to weight saving vs conservative approaches. For example, we conservatively say the caps handle the bending, the web handles the shear, the skin handles the torque. We design for Compression on the caps. (please correct this if it is wrong). But of course this is not entirely accurate, the ribs and skin and carry-through as a system all contribute to portions of those stress. Does the FEA look at all those aspects at once as a system? Has there been any comparison, and conclusions drawn? Conversely does looking at sub-systems as Free-Bodies traditionally give approximately the same result.

#### DaveD

##### Well-Known Member
FEA analyses the model you provide it with, which is both its strength and its weakness. It certainly can produce more accurate results than some more traditional methods especially for complex components that are difficult to model using traditional methods. However the old computer saying "Garbage in, garbage out" definitely applies and a poorly meshed or incorrectly defined FEA model can give completely misleading results. It always pays to do a sanity check of FEA results against some simplified hand calculations. And for complex safety critical FEA applications the results are always extensively validated against real test data in order to "refine" the model (I.e. put in some fudge factors!)

Also FEA is definitely an analysis tool not a design tool. The amount of labour involved means that you usually develop your structural concepts using traditional methods to give an indication of what will and won't work, and then use FEA as an optimisation tool to fine tune the design (shed that last few pounds or get a bit of extra stiffness). Also much like the manual calculations, FEA won't find something you're not looking for... So you'd need to set up an analysis to look for, say, global buckling failure. You can't just model the part once and expect FEA to tell you that a panel will fail in crippling, or the skin will fail in compression.

I guess my point is that FEA is complimentary to traditional analysis, not a total replacement, but it can certainly do things traditional analysis cannot!

#### proppastie

##### Well-Known Member
Log Member
Also FEA is definitely an analysis tool not a design tool. The amount of labour involved means that you usually develop your structural concepts using traditional methods to give an indication of what will and won't work, and then use FEA as an optimisation tool to fine tune the design (shed that last few pounds or get a bit of extra stiffness). Also much like the manual calculations, FEA won't find something you're not looking for... So you'd need to set up an analysis to look for, say, global buckling failure. You can't just model the part once and expect FEA to tell you that a panel will fail in crippling, or the skin will fail in compression.

I guess my point is that FEA is complimentary to traditional analysis, not a total replacement, but it can certainly do things traditional analysis cannot!
This sort of goes to the heart of the question, given the labor involved is it worth the trouble to try to model the whole system, Right now I am looking at perhaps a check of the hand calculations of the subsystems. Do you have a feel for how much change? 5% 10% ???? Assuming of course a properly modeled and properly hand calculated......(a big assumption at this stage for me)

Or conversely, for a simple aircraft (home-built glider) a properly hand calculated system is just as accurate as FEA.

Last edited:

#### Jay Kempf

##### Curmudgeon in Training (CIT)
Those of you that do FEA, has there been any comparison to weight saving vs conservative approaches. For example, we conservatively say the caps handle the bending, the web handles the shear, the skin handles the torque. We design for Compression on the caps. (please correct this if it is wrong). But of course this is not entirely accurate, the ribs and skin and carry-through as a system all contribute to portions of those stress. Does the FEA look at all those aspects at once as a system? Has there been any comparison, and conclusions drawn? Conversely does looking at sub-systems as Free-Bodies traditionally give approximately the same result.
FEA can verify what you suspect might be a complex load path where you can estimate at what level of stress and strain a part will fail and it will give you a location. However, you have to have a very good understanding of the methods, theory, and assumptions to allow yourself to think that you actually have the answer you want. For a one off where you may not be destructively testing a bunch of samples with varying tolerances of the critical areas and varying tolerances of material science and especially with composites you would be being pretty dangerous to rely on a part with a low factor of safety. So in the end what did you lean from FEA? Well, you can get an idea where the loads will concentrate the stresses and in what pattern so that if you believe your assumptions you can rearrange the structure to optimize it. So for that it is worth it. In the end you would then back calculate those hot spots by hand to verify that the FEA was giving the right value which isn't all that hard normally and having gone through that you can get some confidence that the lesser loaded and stressed areas are going to not be a critical part of the investigation. I see FEA as a great optimization tool to investigate configurations of complicated structures with balanced forces. I haven't done it but reinforced concrete is a great example of something that could be investigated with FEA. In that world compressive and tensile forces have to be balanced just like in composites.

An answer to your last part. Yes, FEA can look at balancing a structure with skin and spar contributing to a whole. The problem get's larger when you do that. Looking at individual sub assemblies works very well to limit the problem if you track your assumptions it can be very accurate. When you have bonded structures evaluating parts individually can be dangerous especially when you have vastly different tensile strengths and moduli bonded together.

#### DaveD

##### Well-Known Member
This sort of goes to the heart of the question, given the labor involved is it worth the trouble to try to model the whole system, Right now I am looking at perhaps a check of the hand calculations of the subsystems. Do you have a feel for how much change? 5% 10% ???? Assuming of course a properly modeled and properly hand calculated......(a big assumption at this stage for me)

Or conversely, for a simple aircraft (home-built glider) a properly hand calculated system is just as accurate as FEA.
Serious FEA, especially for composites, requires expensive software and a very high level of understanding & expertise. So, if you want to model somthing relatively straightforward in Aluminum, then sure, download Z88 for free, lose a weekend or two learning how to use it and see how you go...I did this, and it was quite good fun - but then I'm a bit strange!!! However if you want to model a whole composite wing/spar assembly you are looking at BIG $$for the software (plus a powerful PC) and similarly big$$$for the expertise (either an expensive consultant or months of your own time and money for the required training) and even then, without testing, how confident can you really be in the results (so potentially more big$ on testing).

For a simple one off aircraft the cost of extensive FEA is hard to justify, unless money is no object or you are already an FEA expert with access to the tools. If your design is so cutting edge that there is no other way to analyse it or you really, really, really need to squeeze out every last ounce/knot (or, of course, you plan to sell thousands so a highly optimised design will save you money in the long run), then its time to get your wallet out!

IMHO, for a one off, you could spend the $20,000 to$200,000 you'd need to spend on FEA on simply building everything in CF/titanium and shave 30% of the weight!

#### Himat

##### Well-Known Member
The accuracy of a finite element analysis first depends on how accurate and how well the model represents the real thing and the accuracy of the material parameters. Then the mesh (how the model is represented with a number of connected nodes) must be “right”. Different solverers, that is the math algorithm used by the computer, might change the results too. It does take a person or a team with knowledge, skill and experience to get this right. And they need time.

I do use Ansys, a FEA software at my day job, but not anything related to aircraft. From what I have read a structure can be made lighter by using FEA. But it does take someone like Airbus to do so. I do think Airbus managed to design the wing on one of their latest airplanes to have no excess safety margin and then weight. When testing the wing failed at 1.49 something times design load where the specification said it should withstand 1.5 times the design load.

##### Well-Known Member
This sort of goes to the heart of the question, given the labor involved is it worth the trouble to try to model the whole system, Right now I am looking at perhaps a check of the hand calculations of the subsystems. Do you have a feel for how much change? 5% 10% ???? Assuming of course a properly modeled and properly hand calculated......(a big assumption at this stage for me)

Or conversely, for a simple aircraft (home-built glider) a properly hand calculated system is just as accurate as FEA.
For complex structures FEA is the only way to go. But then, why design complex structures in the first place? Even my Warren-truss design is done "by hand" (through automated scripts). Spanwise numerical integration; about 12 different failure modes times a couple dozen "check points" per spanwise section....

FEA for composites is IMHO way beyond the capabilities of anybody, save a specialist with a few years of composite FEA under his belt. Inter-laminar sheer, negative poisson ratio's, moduli that change in direction and per layer...

My only complex part in my own design (tail boom attachment) I plan to validate by testing to destruction. Anything else to 120% of LL or so and validate load calcs by testing a small section to destruction.

#### PTAirco

##### Well-Known Member
I have to agree that analyzing a typical composite structure with FEA is educated guessing at best, unless you have some seriously smart people and expensive software to rely on. In the end the only thing that will reveal how close you came, is by testing things to destruction. And then the practical aspects come into it; the skins have to be able to withstand a certain amount of man-handling, for example and may end up heavier than they need to be from a purely aerodynamic perspective.

On my first biplane project, I did all the calculations by hand, 1930s style. (The different load cases and the complex framework of a biplane are enough to drive you nuts by the way...) Then I got hold of a demo version of NASTRAN and ran the structure through it (frameworks are relatively easy, compared to shell structures) and I was pleasantly surprised that my basic (crude) and conservative analysis came very close to the NASTRAN loads.

In the end a combination of the two seems to be the best compromise, as it will show up any glaring errors in either.

#### cheapracer

##### Well-Known Member
Log Member
Luckily there is a real world comparison to be made in the car world.

Kit cars or custom one off creations require a beam torsion and deflection test to get licence approval in Australia and some other countries and quite a number of FEA simulations have been done during the build then off to get the real world test done by an approved engineer.

A couple of things have shown up, only the top level FEA experts who do it for a living get it spot on. Next is experienced FEA amateurs who generally are within 10 - 20% and then there's the rest who are all over the place, I have seen the same chassis get results of 300% variation.

Of course even having the exact results still doesn't help if you don't know how they apply to your application.

#### proppastie

##### Well-Known Member
Log Member
Next is experienced FEA amateurs who generally are within 10 - 20% and then there's the rest who are all over the place, I have seen the same chassis get results of 300% variation.

Of course even having the exact results still doesn't help if you don't know how they apply to your application.
Great answer, sort of what I was looking for, The experienced FEA amateurs and rest of us,......do you know what software they generally use,.....it probably is not the same as the professionals, but if it is something the rest of us might have access to do you know what they used?

#### Matt G.

##### Well-Known Member
In industry, Nastran is probably the most common solver, using either Femap or Patran as a pre/post-processor, followed by Abaqus, all of which I have used in my job. You can go out and buy a seat of one of these, but it's extremely cost-prohibitive.

I have been playing around with couple of freeware solvers (z88 Aurora and Lisa) and so far I have not been impressed. Both, particularly z88 Aurora, have really clunky user interfaces, and the element types you can pick from are pretty limited. The free version of Lisa has limits on number of nodes and elements. Applying loads representative of aero loads is nearly impossible without oversimplification of the loads, as far as I can tell. That there will erase any benefit of trying to get as much weight as possible out of your structure. There are others that are supposed to be really powerful but don't have much of anything for a GUI; I've avoided these because I have no desire to make FEM models like it's 1984. It's very difficult to perform checks on your model when you can't see it. Plus, I'm not sure I really trust that the solvers are as robust as the commercial codes. Even today there are issues with the commercial codes that crop up once in awhile. The built-in FEA analysis in CAD software is only good for pretty pictures, nothing more.

For what you are trying to do, I would not waste time making a FEM. No matter what kind of analysis you do, it needs to be validated by testing anyway. It takes a significant amount of time for an experienced person to make one of these models, and it will take even longer for a beginner, and you may get done with all of it and get completely bogus results. I also wouldn't recommend a beginner trying to design something so highly optimized that everything is driven to zero margin with the least conservative analysis possible. Even in industry this is generally not done.

#### proppastie

##### Well-Known Member
Log Member
For what you are trying to do, I would not waste time making a FEM. No matter what kind of analysis you do, it needs to be validated by testing anyway. It takes a significant amount of time for an experienced person to make one of these models, and it will take even longer for a beginner, and you may get done with all of it and get completely bogus results. I also wouldn't recommend a beginner trying to design something so highly optimized that everything is driven to zero margin with the least conservative analysis possible. Even in industry this is generally not done.
Its difficult to respond, it is a difficult subject. A study of Buhrn, Perry, Douglas, EAA/Wilford, reveals many uncertainties to the inexperienced. The same with Lisa, and Autocad Inventor. I guess the plan could be to try to make the text books and Douglas design guide agree with the Wilford, EAA, spread sheet, and perhaps agree with the Inventor, (or solid works I can get) along with perhaps any self produced spread sheets. Then of course there is the loading proof of test to Utility limit, along with measurement of deflections. As to a waste of time,....I have not had so much fun in a long time. You guys have been very patient, and extremely helpful, Thanks all.

#### Matt G.

##### Well-Known Member
A study of Buhrn, Perry, Douglas, EAA/Wilford, reveals many uncertainties to the inexperienced.
There will be 1000x more uncertainties with a FEM. It will be easier for people to help you with hand analysis. It will be nearly impossible for anyone to do so with a FEM.

#### proppastie

##### Well-Known Member
Log Member
There will be 1000x more uncertainties with a FEM. It will be easier for people to help you with hand analysis. It will be nearly impossible for anyone to do so with a FEM.
No argument there, but what if the FEM agrees with the Willford????? Which brings up the question, to those experienced, If I design a spar with the EAA Willford spread sheet, will it be an accurate and safe design? Has there been problems reported?

#### Matt G.

##### Well-Known Member
Depending on the quality of your assumptions, they could both be right, or they could both be wrong. No way to know without doing some testing.

The outcome of designing a spar with someone else's spreadsheet (or a FEM or anything else) will depend on the quality of inputs you provide it and how you interpret the results.

#### skyscooter

##### Well-Known Member
No argument there, but what if the FEM agrees with the Willford????? Which brings up the question, to those experienced, If I design a spar with the EAA Willford spread sheet, will it be an accurate and safe design? Has there been problems reported?
Regarding the EAA spreadsheets, they have the following disclaimer at the top of the page:

***This spreadsheet is for educational purposes only and may contain errors. Any attempt to use the results for actual design purposes are done at the user's own risk ***

It is important to remember that there is no substitute for actual structural testing of a new design. Analysis is done to reduce the risk of failure during testing, not to eliminate the need for it.

#### proppastie

##### Well-Known Member
Log Member
Regarding the EAA spreadsheets, they have the following disclaimer at the top of the page:

***This spreadsheet is for educational purposes only and may contain errors. Any attempt to use the results for actual design purposes are done at the user's own risk ***

It is important to remember that there is no substitute for actual structural testing of a new design. Analysis is done to reduce the risk of failure during testing, not to eliminate the need for it.
absolutely that is part of the plan see post #12.