What boundary condition techniques do you use to keep FEA models realistic?

[EdwardNigma]

Hello all,

I work in the aerospace industry as a stress engineer. I recently took a very solid aerospace FEA course as part of a master’s program, and one thing I really appreciated was how the instructor emphasized connecting numerical approximations to real-world behavior. Around the same time, a technical fellow at work recommended a practical FEA book that walks through the process of developing and validating models—which reinforced the same idea.

One thing that stood out to me is just how important it is to set up realistic boundary conditions. A model can easily become too stiff or too soft if you’re not careful, especially when you’re trying to represent how a structure interfaces with its surroundings. This seems like one of the most critical aspects of getting meaningful results.

That leads me to my question: what boundary condition modeling techniques or rules of thumb do you use to make your models more realistic?

For example:

• In truss-like structures, using a pin on one end and a roller on the other can allow for lateral movement and prevent over-constraining.
• When modeling plates, allowing for lateral deformation can better capture Poisson’s effect.
• In 3D space, the 3-2-1 rule (restraining three points to prevent rigid body motion) seems like a solid starting approach.

If you have experience creating robust and realistic FEA models, I’d really appreciate hearing about any methods or strategies you’ve developed over time to handle boundary conditions effectively. Thanks in advance.

 

[GregLocock]

I hit the structure that it is fastened to with an impact hammer and an accelerometer. That tells me the impedance in the normal direction. I frequently have had to model the test rig as well as the IUT.

 

[SWComposites]

other than the 3 simple examples you list, there aren't any general simple rules. have to look at the actual structural connections, and ensure that the BC's are realistic and appropriate for expected structural behavior. common mistake is to over constrain the model, by using rigid BC's or constraints. and do not assume test fixtures are "rigid".

oh, and ALWAYS plot the model displacements for each load case BEFORE looking at any other results. if the displacements do not look reasonable, the stresses are likely rubbish.

then examine the loads/stresses at the BC's; if there are wild peaks then the model is likely over constrained.

 

[rb1957]

boundary conditions are one thing, the connection between elements another. It is easy to (unrealistically) weld them together; on the other hand using common nodes can greatly simplify a model. And releasing individual freedoms may be analytically correct, but may create only a small difference in the key results (so is it worth the effort ?).

 

[FiniteMatt]

I think testing sensitivity is an important part of the process. How much do your boundary conditions even matter for the study and loads you're applying? Run two studies, one with 100% fixed BCs, and another with some soft springs or some other alternate restraint and see what difference it even makes to your results. Can you live with the results of the highest stressed setup? Or do you need to do a little more digging and quantify the restraint better?

 

[EdwardNigma]

I think testing sensitivity is an important part of the process. How much do your boundary conditions even matter for the study and loads you're applying? Run two studies, one with 100% fixed BCs, and another with some soft springs or some other alternate restraint and see what difference it even makes to your results. Can you live with the results of the highest stressed setup? Or do you need to do a little more digging and quantify the restraint better?

Thanks FiniteMatt. I appreciate the advice as it's what I was looking for. It seems to me that boundary conditions are really important in representing a real life structure. From my class I understood that one needed to perform multiple studies to see what was a more realistic model.

 

[JoshPlumSE]

I think testing sensitivity is an important part of the process. How much do your boundary conditions even matter for the study and loads you're applying? Run two studies, one with 100% fixed BCs, and another with some soft springs or some other alternate restraint and see what difference it even makes to your results. Can you live with the results of the highest stressed setup? Or do you need to do a little more digging and quantify the restraint better?

I'm a little late to the party here, but I wanted to elaborate on this excellent comment. I was the VP of Engineering / Tech Support / Training of a structural engineering / FEM program for a number of years.

Here's what I used to tell our users about the Truss example that the OP mentioned. People would sometimes model Pinned - Pinned instead of pinned roller because the truss was supported on either end by a concrete wall or something else relatively rigid. I'd tell them to do the following:
a) Model it both ways.
b) The pinned - pinned model will produce a large reaction at the location where the the roller exists in the other model.
c) The pinned - roller model will produce a deflection (but not reaction) at that same location.
d) Compare the two. If the reaction is 200 kips and the deflection is 0.01 inches, then ask yourself what you think would happen if you applied that 200 kip force to your supposedly rigid wall? If it would deflect 0.01 inches or more, then you can be completely confident that the pinned - roller model is more accurate.

 

[rb1957]

With the above example, I'd've said instead of a pinned-pinned model, which creates an infinitely stiff support with no deflection along the length of the beam, use a finite stiffness ... a CBUSH or a beam element (to the rigid constraint). Then again you can see the influence of allowing small deflections, which should be very reasonably realistic real world things, to happen and assess.

 

[GregLocock]

That is in the pinned/roller example is there any axial motion of the roller? My homebrewed truss program says yes there is axial motion, hence unlike the simple hand calc there is a difference between pinned pinned and pinned roller.

Since we can use superposition Pinned pinned=pinned roller+ a stretching force at each constraint

Well that's odd. Consider a pinned roller truss symmetrical about the X axis. Applying a force stretching it along the x axis will produce no vertical deflection along the center line. A uniformly distributed load normal to the x axis will produce a deflection of the centre line. Yet it is intuitively obvious that pinned pinned will introduce an x reaction force at each constraint, and will have smaller deflections than the pinned roller case.

Um, time to play about.