2D vs 3D analysis

[ThomasH]

Hi all

I have a few short questions in connection with a discussion I've had with collegues recently:

1. Given a typical 2d problem, would you use 2d or 3d software to solve it?

2. Given a 3d problem, would you use 3d software or divide the problem into 2d subsystems and use 2d software?

3. Would you in a broader perspective consider 2d-analysis as "generally ok" or as "outdated"?

Any other ideas on 2d vs 3d analysis?

Thomas

 

[prex]

My answers:
First of all I assume that referring to 3D you mean solid elements: a 3D model done with 2D elements (shell) is not considered 3D here.
1)Would of course stay with 2D: no reason to go to a more complex analysis if the problem doesn't require that
2)Personally consider that true 3D problems (those that can not be solved with 2D analyses) are very rare: I would all the time start with 2D analyses, that would form the basis for checking the validity of 3D results
3)Already responded to in 2)
3D analyses are very hard to check, as true 3D behaviors are difficult to be represented in our minds, and thus to be evaluated for realism.
Also consider that a few stress and strain phenomena (we are speaking of structural analyses, don't we?) require a true 3D analysis: fatigue and transient thermal stress are examples that come to mind, but anyway under quite special circumstances.

prex

http://www.xcalcs.com

Online tools for structural design

See

http://www.xcalcs.com/docs/symbolinfo.htm

if you want to use symbols on these fora

 

[corus]

1. If it was a 2D problem then you use 2D siftware, obviously

2. 3D problems can be approxinated by 2D analyses providing you know the limitations of your analysis. A thick section can be aproximated by 2D plane strain providing there is no out of plane loading, similarly for thin sections where plane stress can approximate the stress distribution. In this case the plane stress approximation will be less valid internal to the section, and the thicker the section the less valid the results. Similarly axisymmetric analysis can be made if the profile of the swept path is not quite symmetric abotu the axis. In all cases it is better to check the validity of your 2D assumption on a 3D model with test load cases. With present day computing power, though, it's easier just to do a 3D analysis if there's any doubt in your mind.

3. 2D analyses are not outdated, similarly 1D analyses (such as in heat flow) can produce valid results providing your assumptions are correct.

corus

 

[ThomasH]

Hi again

Thank you prex.

Just to complicate things I want to limit the question to the use of 1d and 2d elements, beams and shells. But it was very interresting to hear your distinction between 2d and 3d analysis.

My questions can be formulated in another way: Would you consider 3d analysis (one model) more efficient than 2d analysis (more than one but simpler models)? In this scenario you would have to move the 2d data between models "by-hand".

Regards

Thomas

 

[prex]

If you limit your question to 1D and 2D elements and the problem is truly three dimensional (e.g. 3D trusses or domes with non axisymmetric shape), then I don't see why you shouldn't use a 3D model.
My remarks were mainly on the difficult interpretation of stress distributions in a three dimensional continuum, no problem for interpreting instead 1D or 2D stress distributions, irrespective of model dimensionality.

prex

http://www.xcalcs.com

Online tools for structural design

See

http://www.xcalcs.com/docs/symbolinfo.htm

if you want to use symbols on these fora

 

[imagitec]

"My questions can be formulated in another way: Would you consider 3d analysis (one model) more efficient than 2d analysis (more than one but simpler models)? In this scenario you would have to move the 2d data between models "by-hand"."

I'm a novice, but I've already learned that automatic meshing is far from the most important thing when it comes to getting useful and reasonably accurate data from an FEA. Systems such as CosomosWorks that don't permit manual meshing can even prevent you from getting a solution, if they fail to mesh, let alone doing so efficiently.

Rob Campbell

 

[rapt]

If the problem can be relatively accurately treated as a series of 2D members, this will be far quicker. I can define, analyse and design and evaluate the results for a post-tensioned slab or beam in 2-3 minutes using my 2D software. And everything is there for me to verify the results easily. So a normal floor might require 5 to 10 runs or 10 to 30 minutes. This is not normally possible in 3D software.

If the system is complicated and very non-symmetric and hard to break down into logical 2D members for analysis and design I would consider 3D analysis but may do some 2D as checks.
If automated design was also required I would only use 3D software to do this if I was very experienced in the design area and I would definitely do 2D check runs and check the overall results carefully. This type of design hides a lot of what it does from the designer and the designer often has to input a lot of the design requirements for different areas of the problem. Making sure that the whole design has been completed and that the results are ok is often not easy.
A reinforcement drawing is not a good set of results to check. A good designer requires more background numbers. These are much easier to access in good 2D software.

 

[bvi]

I think it would help in providing specific answers if you described the structure or component you are considering for analysis.

In general, one should use the analysis technique that captures the essential features of the system being analysed. 1-D, 2-D and 3-D can all be appropriate.

Although this is not the question, with respect to the issue of modelling shells with 2-D shell elements versus solid elements, it is my experience that the shell elements provide more accurate results and are also easier to use in constructing the model.

With respect to evaluating a structure by pieces, and adding the pieces together by transferring data from one analysis to another by hand, this is a matter of judgement and resources. However, the trend is of course to model the larger structure as a whole, because it is cheaper and more accurate. The computer power has increased to permit this, potentially inaccurate simplifications are avoided, the time it takes to transfer data from one model to another costs money, and the data transfer introduces inaccuracies. I remember one project I reviewed in which the coordinate directions for the loads transfered from one model to another model (the structure was split into partial models)got mixed up.

 

[imagitec]

ThomasH:

"Would you consider 3d analysis (one model) more efficient than 2d analysis (more than one but simpler models)? In this scenario you would have to move the 2d data between models "by-hand"."

cb4 wrote:

"With respect to evaluating a structure by pieces, and adding the pieces together by transferring data from one analysis to another by hand..."

I don't think ThomasH was referring to contriubtion versus assembly analysis. I believe the two models he was referring to were the CAD model and the FEA model. 2D certainly doesn't rule out assembly analysis, contact, etc.

Even the idea of a single model when using an automatic, solid mesher is misleading. A mesh is a mesh, not an ACIS or Parasolid model. Today's embedded applications store their data in the CAD file, but it's not the same data. They are very good at maintaining associativity between corresponding model surfaces and parts of the mesh so, for instance, loads don't need to be reapplied if a dimension is changed (external (non-embedded) tools can have good associativity, too; all-in-one solutions have their advantages and disadvantages).

There are also tools that ease the creation of 2D meshes from models, and there is no reason they couldn't be as associative as the 3d meshes. Are there any packages that achieve this today?

Thanks,
Rob Campbell

 

[imagitec]

I wrote:

"2D certainly doesn't rule out assembly analysis, contact, etc."

Of course it does if the problem couldn't be reduced to 2D.

 

[ThomasH]

Hi again

Thank you for your input.

My original idea was to compare 2d and 3d analysis without any predjudice. For that reason the question wasn't exactly clear.

For example a simple portal frame can be analysed with both 2d and 3d software. The reason for using 3d software would then be that you would only need to know one software package.

On the other hand, expand the portal frame to 3d by adding a second frame outside the first one and connect them with beams. The four columns will form a square. This can be analyzed as one 3d model or a couple of 2d models. (Due to symmetry etc one 2d model might suffice but that is not the point.) Would 2d or 3d be considered as most efficient in this case?

It is sometimes said that 2d-software is faster to use than 3d-software but is that always true? In my experience 3d-software often has better modelling tools giving it better performance. This is then based on that you use the software often enough.

Regards

Thomas

 

[bvi]

If the 3-D structure is simply repetative 2-D structures, then it is quicker to do a 2-D analysis, whether you do it with software capable of running only 2-D problems, or software that can run 3-D problems. You need a model that captures the essential elements of the problem. Today, purchasing software, I would advice you to purchase software capable of 3-D problems because not all problems lend themselves to accurate simple 2-D solutions.

 

[jhardy1]

Thomas,

Is 3D better than / more applicable than 2D?

The short answer is yes. And no.

The long answer is:

If the case you are looking at can be well represented by one or more 2D analyses, it would generally be a good idea to start with a 2D model – you can always expand to 3D later, if you need to, but it can be much harder to wind a 3D model back to a 2D model.

The benefit of this approach (start simple, add complexity as required) is that you can do numerous very fast “proving runs” with small simple models, then create more detailed models (if necessary) for final design checking. Large complex models (especially 3D solid models) can be much harder to debug, and take much longer to run, than simple 2D models. The 2D models will generate sensible answers that you can use for “what if” scenarios, and to gain confidence about the results from the more complex models.

Don’t forget that 2D theory doesn’t mean only plane frame type behaviour – it also encompasses plain stress, plain strain, and axi-symmetric theory, so one of these might be very useful even for a problem that you initially think of as being a 3D problem.

If your problem simply can’t be well represented by a 2D model (eg irregular solids), then 2D analysis can only give you a very rough approximation at best.

If you are using 3D capable software, there is absolutely nothing to stop you doing 2D analyses – e.g. simply model everything in the XY plane, and constrain z direction displacement, and rotations about the X and Y axes. If you only have 2D software (eg Plane Frame analysis), then you don’t have the tools you need to consider “true” 3D problems.

For the case you describe of a series of parallel frames – if all loads and displacements are parallel to the planes of the frames, then a 3D analysis will add very little (if any) value over 2D. It will take you a bit more time to model and apply loads and constraints in the 3D model, but the results should be exactly the same as if you had modelled a single 2D frame. However, if there are some loads or displacements perpendicular to the frames, the 3D model would allow you model all behaviours in a single run. For example, a typical portal frame building has several parallel portal frames, plus some longitudinal strutting and bracing. For purely transverse loading, there should be no design actions in the longitudinal strutting and bracing, but for longitudinal loading, these members would be mobilised, and their design effects would be additive to the transverse loading effects. You can do all of this in a single run in a 3D model, but you would need separate 2D models of the portals and the longitudinal bracing if using a 2D package, and you would then need to manually add the design actions from the 2 analyses to get the total member effects.

In 3D models, you can incorporate 1D elements (beams / struts) and 2D elements (plates and shells) together with 3D elements (bricks and test). Just be VERY careful about making sure they are joined in an appropriate fashion to transfer the incompatible displacement and load effects between elements of different types. It’s probably best to avoid this practice unless you are really sure you know what you are doing.

Hope this makes sense

 

[imagitec]

JulianHArdy wrote:

"If you are using 3D capable software, there is absolutely nothing to stop you doing 2D analyses – e.g. simply model everything in the XY plane, and constrain z direction displacement, and rotations about the X and Y axes. If you only have 2D software (eg Plane Frame analysis), then you don’t have the tools you need to consider “true” 3D problems."

I think the choice (and problem) that people will increasingly face is not between 2D and 3D but between manual and automatic meshing. Programs like CosmosWorks only do automatic meshing. People may choose ease of use over capability and only realize the true cost of this later (or never realize it, putting in extra effort or never solving particular problems because they can't mesh).

There is no reason a full-blown package cannot also have the ease-of-use and associativity of typical CAD-integrated packages. But there are business reasons why an FEA company might only want to focus on the latter. The market for these is much larger but also less sophisticated. Better to hide the advanced features that that market might not value; it would only add to the support costs of the company.

Rob Campbell

 

[ThomasH]

Hi again.

Thanks for your input. The original post was based on a comment that "3d analysis is very expensive and generally overkill".

I just wanted to test the forum and see which direction the discussion would take. The conclusion seems to be that 2d might catch the true behavior in some cases (I agree) otherwise 3d might be required (again I agree). The general opinion seem to be that the interresting thing is to model "real behaviour" and that can mean 1-, 2-, 3- or ... analysis.

There has also been questions regarding automatic meshing and FEM or CAD models. This is an interesting area even though it was not what I originally was aiming at. However, in the future with more virtual product development the communication between FEM and CAD will most likely be an issue worth discussion. I think that a FEM software that imports the CAD-geometry and meshes automaticly with no user interaction might look good "on paper" but it has very limited worth. But, the future is something I can only guess about and I have been wrong before.

Best Regards

Thomas