Compression testing pultrusion

[ragflyer]

There was a post above that cited the composite expert at Boeing as stating the test method that produces the highest strength reading on composites in compression is the right one. The reasoning behind this is that compression testing is tough to get good results on - lots of ways to trip low readings that would then drive overbuild.

ok got it. This is exactly why I feel pure compression or shear readings or in general coupon tests are of limited value in composite design. If weight optimization is the goal then testing the actual article or a geometrically scaled one is the only rational way to proceed in my view.

 

[Rob de Bie]

Harder than compression strengths is shear strength of composites. Ugh. You end up falling back on analyticals...

Long time ago I did ASTM D4255B three-rail shear tests, and I do not remember running into difficulties. Except that I was surprised about the non-linearity of the stress-strain curve, all the way from start to failure.

Rob

 

[David L. Downey]

Had not considered glass rebar or similar products, but that is certainly an idea, and the simple straight fiber bar is probably the ticket.

The original tailspring was supplied initially from I think Featherlight and then from someone else. I never have seen one in person but the drawing has them as more of an oval than a squared off rectangle.

having one in hand, it is an obvious pultrusion (rectangular cross section with full radius ends and no mechanically generated surface) but it has always been a darkish amber in color leading me to believe it was pulled using a higher temperature than typical pultrusion epoxy). You could make one by running a high strength glass pultruded round rod through a sanding drum to generate a parallel flat on each face then finishing the full half round edges by hand?

 

[proppastie]

non-linearity of the stress-strain curve,

the stress strain was a curve (arc) rather than a straight line to yield?

 

[wsimpso1]

Long time ago I did ASTM D4255B three-rail shear tests, and I do not remember running into difficulties. Except that I was surprised about the non-linearity of the stress-strain curve, all the way from start to failure.

Rob

Seems to this experienced engineer that non-linear stress-strain plots ARE a difficulty. Non-linearities should be tiny.

When you talk of non-linear behaviour, are you talking big or small? Experimental error, instrument settling and drift, fixturing errors, samples not mature, and other sources all can add noise to your output that are not supported by physics and mechanics...

Billski

 

[wsimpso1]

I'd never thought about the moment of the neutral axis during the bending test but if Billski is using observation of first fiber failure is there any significant movement of the neutral axis due to effective thinning of the test coupon?

With first fiber failure strains on the order of 1%, the neutral axis shift is tiny, so I never worried over that. You could whip out your mechanics of materials text and compute how much it moved. In my case, all of my failures exceeded book values, so I used book values and moved on. I would have gone further had I gotten strengths lower than book values....

 

[ragflyer]

With first fiber failure strains on the order of 1%, the neutral axis shift is tiny, so I never worried over that. You could whip out your mechanics of materials text and compute how much it moved.

Couple of points that may be of interest. Again I do not know for sure if they broadly apply to FG composites. The effect, though, in wood is dramatic.

1. I think the neutral axis move is due to the elastic modules of the compression side changing/ reducing ( curve bends) while the tension side remains the same; in effect they act as two different materials above some stress. That said I think point 2 may be more important.

2. As such I do not think the neutral axis move is the main factor as the difference in stresses is also seen at limit loads (IOW you do not need yielding or fiber failure). For example, a wooden spar compression flange/side resists a significantly higher limit stress than in pure compression- the compressive limit stress in bending is 5300psi and in pure compression is 3530psi for a rectangular spar. This is a big difference. Why? I suspect perhaps the explanation lies in the distribution of stresses in bending (along the cross section). Fibers closer to the NA may provide additional support to the extreme wood fibers in the case of bending . But in pure compression the stresses are more or less uniform (and always compressive) about the cross section and we do not have the mutual support of the fibers in the same way..

3. In practice either way the exact pure compressive stresses not very relevant. You can use what you get in the bending test to design the spar as it is loaded in bending as well. In case of wing skins, fuselage skins and so forth the ultimate compressive stress is almost never reached because buckling is the critical mode.

 

[wsimpso1]

Yes, nice to hear as that his exactly what I did as wel and found them valuablel. Of course there are some potential gotchas worth noting. Just before bending failure on the compression side the material will yield transferring more load to the tension side resulting in much higher failure strength. In effect the neutral axis is moving towards the tension side and delaying failure. In materials that yield and have a significant delta between tension and compression strength you have to be careful as the compression failure values you get could be higher (not conservative ) than if you did a pure compression test in the short column range. Now this may not be a big deal in highly elastic material like carbon that does not yield. But wood is a great example where this matters.

If mine had done that, I suppose they would be the gotchas. I ramped up load slowly by turning a screw, measuring load and deflection at each increment, and then I would get the first creak and crack sounds. Load-deflection plot showed the curve to go non-linear at first noises, peak at or only slightly above first noises, and so the post yield behavior was interpreted as failure load.

With four point bending I had a couple square inches of surface per sample per side that is both at max tensile or compression stress and at zero shear stress, so this scheme is particularly good at isolating tensile and compressive stresses. I was fortunate in that the white spots on samples indicating fiber failures occurred away from the support points, so loading at the failure points was pure compression.

Three point bending has a positively tiny area at max tensile stress, while the compression side has multiple issues:

• Max compression also sees only a tiny area at max compression stress;
• There is compression in the z axis from the fulcrum that both compresses and constrains that surface;
• The shear carried by the beam goes from one-half of the applied load up to one-half load down at the fulcrum;
• The load situation at the fulcrum then becomes one with high compression in x direction, compression in z direction, and a poorly defined shear situation where shear is changing state.

I did not even attempt to do three-point bending tests as it looked to be difficult to resolve shear and longitudenal stresses.

The issue of resolving stress state and failure modes also comes up in surrogate tests. Particularly with beams, the stress state in the web includes axial loading from caps, so resolving failure state is "difficult". Now if you build a cap of UNI and wrap it with modest thickness of +/-45 in same material, you can compute the fractions of bending stiffness of each lamina and resolve bending stress in the extreme fibers of the UNI. If stress at first fiber failures are similar with simple laminate and wrapped laminate, you can trust your numbers. I also suspect that one could design a UNI laminate cross section with an intended higher tensile than compression stress or vice-versa to precipitate failure on the other side of the sample.

Then you get to shear strength. Our shear webs only see pure shear at the neutral axis. Everywhere else they see shear plus either tension or compression as the section deforms under bending. Computation of failure criteria in the shear webs thus requires us knowing shear and axial strains and stresses at failure of the web laminates. We can easily get axial failure strengths but shear is a tough one to measure. Way back in graduate school we were given some approximate methods based upon netting analysis and others. I do not know if that has been seriously eclipsed, but would love to know if it has. The really strange thing is that we end up beefing the caps to make the web survive, so the caps end up overbuilt. I suppose one could search the design space for a structure that would overload the webs and give us a clean failure point for our shear materials. Me? I used rules of thumb for a shear strength to put through the failure criteria KNOWING that the axial stresses near top and bottom of the web are big, then threw in another pair of crossed plies for Grandma. Not terribly satisfying, but I know it is strong enough and satisfyingly light.

Billski

 

[Rob de Bie]

the stress strain was a curve (arc) rather than a straight line to yield?

From memory, 25 years ago, no notes saved: yes, and substantially non-linear. I had 'tested' the test method with 2024 before, and IIRC those results were as expected. I remember shear modulus to be within a few percent of the listed value. Please note: these were not certification tests, it was more learning about material behaviour.

Rob

 

[Rob de Bie]

Seems to this experienced engineer that non-linear stress-strain plots ARE a difficulty. Non-linearities should be tiny.

When you talk of non-linear behaviour, are you talking big or small? Experimental error, instrument settling and drift, fixturing errors, samples not mature, and other sources all can add noise to your output that are not supported by physics and mechanics...

Billski

Oh yes, non-linear stress-strain plots are a problem if everything you learned revolves around linear stress-strain behaviour! As stated above, I had 'tested' (calibrated ?) the test method with 2024 before, and it worked well. Maybe I made a mistake, who knows, but I do not remember noting any in the data (like slippage), and I'm quite meticulous.

I seem to remember (25 years, no notes, etc) that afterwards I looked into the non-linear behaviour, and found that non-linear stress-strain plots are standard for shear tests composites. I can't back that statement up with sources, it's too long ago. I'm just reporting what I remember from those interesting tests.

Rob

 

[wsimpso1]

Oh yes, non-linear stress-strain plots are a problem if everything you learned revolves around linear stress-strain behaviour! As stated above, I had 'tested' (calibrated ?) the test method with 2024 before, and it worked well. Maybe I made a mistake, who knows, but I do not remember noting any in the data (like slippage), and I'm quite meticulous.

I seem to remember (25 years, no notes, etc) that afterwards I looked into the non-linear behaviour, and found that non-linear stress-strain plots are standard for shear tests composites. I can't back that statement up with sources, it's too long ago. I'm just reporting what I remember from those interesting tests.

Rob

That still sounds like a difficulty with the testing...

Three Points:

• Structural fibers by themselves ARE quite linear elastically;
• Elastic behaviour of composites made of glass or carbon fibers glued together is dominated by the fibers, with the matrix largely going along for the ride;
• These composite materials must then be intrinsically linear. If tests show non-linear, something other than the intrinsic characteristics are being displayed. We had better have some understanding of why or our numbers can be very misleading.

Back in graduate school we had a lab demonstration where linear and non-linear elastic behavior was demonstrated on the same Instron with samples of the same material. The difference was in the sample size and sample prep. Long thin test sections with gradual tapers and well distributed loads gave nicely linear stress strain curves. The professor's message was that when we see non-linear behaviour, we had better figure out how that happened or we will be misled by the results. Short thick sections with poorly distributed loads gave non-linear behavior and lower modulus in general. The reasons given for this were several, with the most easily grasped and repeated one being that it takes roughly 1000 diameters of a fiber for it to fully pick up load from the matrix. The fiber below it takes the same 1000 diameters, but it is offset down stream, repeat through the stack. If the sample is thick in the direction of the gripper or short in the load application direction or both, load is disproportionately loaded to the free surfaces, and you end up reading transitions not intrinsic material behaviour. The rest of this particular description is that most of our composite structures are fairly thin and long and loads get well distributed. Short thick things made of composites, where they exist, will have non-linear behavior while most things that have large transitions and thin sections behave linearly. You can fool yourself badly on characteristics if your test articles do not have enough similarity to the real articles.

Then there are issues with gripper design, immature samples, poor bonds, the list goes on and on, and all defects showing up as non-linear stress strain curves that do not show up in most of our airplane parts.

In the end, if your structure has lots of room to get load in and out of the fibers, it will behave like the samples that do the same. Think spar caps with long continuous fibers and gradual thickness changes or wing skins with well distributed airloads being fed in. If your structure gets loads in and out of the fibers in short distances or through abrupt section changes, we can expect behavior like seen with short rapidly changing sections on test. Think about a long rod bonded into short fittings. Close to the rod ends, the loads are poorly distributed, while away from the rod ends, loads are vary nicely distributed. If the rods are long, a big part of the rod behaves linearly, but a really short stout rod might have messed up load distributions and look non-linear even worse than your test sample.

In beams, I tend to believe that caps being long uni structures, loads are well distributed with the full linear modulus in play. So are wing and fuselage skins. Go to shear webs in beams, and the fibers run diagonally between caps and lap onto the caps, they might have some non-linear behaviour in the radius where they bend from cap to web proper and pick up shear loads. Over the rest of the web, the jury is out in my mind.

I do know that wing deflections measured in tests, once settled in their test rig, tend to look linear. Nice straight lines relating load and deflections... In cases with large angular deflections and using vertical loading the structures ten to show stiffening as the load goes up, and that is simply geometry of the test, not because the material gets stiffer. The main non-linear areas of the plots are in the low load end of the range, which is mostly settling the rig. All of this is both common and encouraging of linear assumptions. With that knowledge, where do we see any sense in using test data that shows non-linear behaviour? Now if your systems actually show increasing deflection vs load slopes, and in proportions to the sample test data, more power to you. Both my experience and the experience of the folks who trained me in this topic did not reflect that sort of thing, and made me skeptical...

The rest of the story on composites is that we use factor of safety of 2.0 and more in composites, tend to overbuild skins (working at min gauge for surviving build, ground handling, and airshow morons), overbuild spar caps (The light way to strength in spars is overbuilt caps, and reducing deflection also drives overbuild), and even the shear webs tend to end up overbuilt. In the end, these structures are living in the first third (or less) of the stress strain curve, which, even with a lot of short fiber and poor load distribution effects still looks close to linear. So we can still use those E's and Q's for the lower third of the stress strain curves and linear theory, even if they are somewhat sketchy.

I like having a rational path from believable data. The thought of using non-linear data that may reflect defects in testing more than actual part behavior is spooky beyond my tolerance. I would hope it is to you guys too.

Billski