Fatigue of Riveted Connections 6

[JohnnyEnglish]

A conservative method of predicting the fatigue life of a riveted joint is to extract nodal forces from an elastic finite element model with the rivets represented as beams, calculate the maximum sheet bearing stress using the quotient of highest load and rivet bearing area on the thinnest sheet, and then deriving the number of cycles to failure from the unnotched material S-N curve.

Any thoughts?

(Apologies for the double post)

 

[edbgtr]

Johnny
Not aware of the "rivets as beams" bit, usually use effective spring constants obtained from appendices in Niu's pink book, or from Bill McCombs' Bruhn Supplement, Appendix D, available soon.
There has been quite a bit of discussion on this forum regarding fatigue of rivetted joints and I suggest you search around for this valuable information. The British have done quite a bit of work on this subject, as has Prof Schijve in the Netherlands. A good source of reference is the ESDU data sheets, and some of the original work done by Schijve which can be found on the NACA site at

http://naca.larc.nasa.gov/

There are secondary effects, like clamp up that can change the life dramatically, so careful note should be taken of the detail in your particular application before assessing the life of the joint. In fatigue "the devil is in the detail" and failure to adhere to the finish "rules" can cost you dearly.
Ed.

 

[JohnnyEnglish]

Ed,

Thanks, for replying; this stuff must be fairly proprietary info as it seems to be a well guarded secret...aerospace companies must do it all the time. Being part of a small company means that we do not have the resources or funding to conduct significant suites of experimental testing on actual joint configurations as the Boeings etc of this world must have done...(hopefully).

I actually posted a more complete description of my problem in the FEA forum (I was hoping to generate some discussion here in this forum amongst those who know...). I'm familiar with the etiquette of these databases (databasi?) and I did search and dig and delve into all sorts of avenues. I have used Niu's modified stiffness formulae in simple models and tried to reproduce his load distribution for the lap joint configurations which he gives as examples, but they appear to be wrong. Not only that, but his example of how to predict the Severity Factor is dubious, and rather conveniently comes out to exactly 4...

Since posting here I have had some success - I modelled a very basic example from Shigley, (ex. 8-6), at first with couples instead of beams for the rivets/bolts and the forces on the fasteners were incorrect. Shigley's example by the way quotes the lower force to be 13.8N when it should be 14.8N, which threw me for a while. When I modelled the rivets as beams with regular stiffness and the joint members with stiff shells offset from one another by a material thickness, the loads were as theory predicts (which is at least a start). Then I put the connected members to their regular stiffness (193 GPa) and got a slightly different, and hopefully "real-life" correct load distribution. I didn't model the clamping effects and I know these improve the life of the joint, but I'm looking for conservativeness (hmmm).

The reason I'm using FEA is that the joints I'm modelling are part of a complex structure of riveted members and are often not straight forward lap joints in shear (as an aside - what's the rule of thumb for max tensile load on a rivet...I seem to recall 25% of max shear, but I can't be sure). It's also a random vibration analysis (psd) to help predict survivability of an accelerated vibration test (mil std 810) from which Ansys gives me a statistical value which I can use with Miner's rule and the S-N curve of the material to predict the number of cycles to failure, but I won't go into that. However, I need to know if I can use the bearing stress which I described above to get that first statistical value from which all else will flow.

Tied in with the vibration problem is the prediction of fundamental frequencies using FEA with a riveted structure. Does anyone have any experience in this area? Couples seemed better than beams, but like I said, give incorrect load distributions as they act as stiff connections. I'm pusing to get some simple lap joints sine swept to compare with models at the moment. As for going non-linear and into fracture mechanics....that might make my head explode, and no-one wants that.

Sorry it's so long, but I had to vent.

JE.

 

[jetmaker]

JE,

The general rule for rivets that I am aware of is not more than 10% of their rated shear capacity in tension.

regards,

jetmaker

 

[JohnnyEnglish]

Jetmaker - any references? (legally I'd be pushing it with ref: "jetmaker from eng-tips", although I've gotten away with a lot in my time ;-)).

Thanks,

JE

 

[jetmaker]

Funny JE...

Well shut my mouth... I have 2 references for you, and they certainly do NOT agree with the 10%.

Ref. 1) Jean-Claude Flabel, Practical Stress Analysis for Design Engineers, pp. 149. Suggests a rule of 2/3*Psu for protuding head rivets.

Ref. 2) Bruhn, pp. D1.25. Suggests various criteria and methods, PLUS... provides tension test data in tabular form for various rivet types. Tables show tension strength as a function of thickness.

There are some refs. for you. The use of the 10% rule seems to be conservative in all conditions.

Hope this was helpful.

jetmaker

 

[edbgtr]

Hi JE
Getting into a general rivets debate could sap your energies and deflect you from your original goal. Try not to use the darn things in tension, they're not made for that. Use bolts or the "specials", if your application demands it.
Niu's SF method is not his, but was developed by Jarval of SAAB, in an attempt to cater for the huge scatter in rivet fatigue data. May I also suggest that you take a look at the fatigue sections of the British Def-Stan 00970 (both for helicopters and fixed wing aircraft, Chapter 200, as I recall), also available for download on the Internet, for their handling of wide-band S-N data, and "unknown" load spectra, in fatigue calculations. I have used it and get surprisingly good correlation, when used in conjunction with known aircraft fatigue spectra manipulated to yield pseudo-range-pair load cycles. The "Niu" SF method is used in VERY reputable aircraft company structures manuals, so don't dismiss it out of hand. Your use of "beams" to model the rivets should yield the moment condition (within the attached sheets, i.e. your shells) which adds to the local bearing load, and hence one of the SF terms used by Niu. As a matter of interest, if you have the classical "Socket Analysis" data sheets from a "manual", you can unravel this to produce the same results as the moment created by the off-set load in a joint. This moment can be reduced by the moment reacted under the fastener head produced by clamping of the joint. This however puts the rivet into a complex tension and shear condition and it's failure criterion changes.
Regarding natural frequencies; when analysing (or testing) the natural frequencies of a riveted assembly, you should detect a damping term (damped natural frequency) caused by the friction effect between the faying surfaces. So, if your FEA model can simulate, by means of friction contact elements between your shells, you may be able to determine the damped frequencies of the structure. This could, indeed, be head exploding stuff .
The ESDU data is not company proprietary material, and can be purchased from the ESDU in England, at a price mind you. Use Google to find them.
Good luck.
Ed.

 

[edbgtr]

Johnny
If you are still interested in this problem, I have discovered some work by Thomas R Brussat, PhD, on the subject of FEM analysis and changing fastener loads with crack length.

This work was presented at an Aging Aircraft Symposium in 1990 held in Atlanta, Georgia, where Tom Swift also presented some very good charts on fatigue life of shear lap joints. This may help you to calibrate the work you are doing.

Ed.

 

[crackman]

Hi Guys

I've been playing, which is all I can really call it, with fatigue methods, SSFs, spring/splice models for more than 15 years now. The best and most accurate way that I have found to obtain a somewhat reasonable (note, fatigue is nothing if not stochastic) fatigue life prediction methodology is to use actual joint SN data. If one searches hard enough and thoroughly enough, there are literally dozens and dozens of public domain test data reports/papers on joint SN testing.

First off, the NACA and wartime reports(yes, WW2) have an over abundance of both riveted, bolted, countersunk (even over 100%), dimpled, etc. joint test results. In addition, many OEMs published several technical papers in the 50's and 60's with great referenceable data on joint testing, they are just hard to locate. I believe many of the old ICAF papers have such information as well. I myself have a collection of over 100 referenceable reports which has taken quite some time to locate over the years, but it can be done.

Now, since I work for an OEM, I have the added luxury of adding more test data and I was able to develop my own alpha and beta factors to adjust my SSF to account for various effects. However, this is not entirely necessary as you could use the actual joint SN curves themselves and then factor them. Main point is, the Mil-Hndbk 5 SN curves themselves do not accurately predict joint fatigue lives. I have tested hundreds of joint specimens and generally if you use MH5 it is very conservative at high max stresses but very unconservative at lower stresses.

Note that effects such as fastener fit, countersink, finish, hand vs machine driving, deburring, etc. can far outway any amount of "polishing the beebee" in building nice detailed splice models, I know, I have built my share of them. Just as a side note, I have investigated blind fasteners as well and have performed many actual tests on them. A blind rivet in a machine countersunk hole can have a life reduction factor of over 2 compared to a hole filling rivet in the same machine countersunk hole. Also, a dimpled rivet has twice the life of one in a machine countersunk hole. Another surprising effect is bucked tail size. The life can actually vary as much as by a factor of 3 to 4 between a low and medium sized bucked tail. This reflects the effects of clamp-up and the resulting friction effects.

So, the lesson is, see the forest for the trees. Dont spend too much time predicting the % load transfer to a gnats tail and then overlook obvious more important impacts. Make a good conservative determination of the load transfer and move on and make sure your method accounts for the other effects.

James

 

[edbgtr]

James
Would you consider making a scanned, .PDF set of your referenceable data available to others not quite as fortunate as you?
Confirming, or denying, analysis techniques is the name of the game with this stuff.
It would be greatly appreciated.
Ed.

 

[crackman]

Ed

I myself would love to have all of my references scanned, but I am afraid it would take months if not a year to scan them all. I literally have a bookcase full of them. I am currently working on a Airframe Fatigue Handbook. Purely a practical application book containting a lot of historical perspectives, public domain references and practical methods and criteria for performing fatigue and damage tolerance analyses of basic airframe components. I will be including a good section on spectrum development as well. But the completion of that endevour is off some time in the future.

I would be glad to put together a list of publicly available references though that I think are worth while for any one interested. I will look thru all of my references and pick the ones out that are most helpfuly. By the way, I have picked up some great material (less than $10) off of ebay that I had been searching for for quite some time. So keep your eyes out on it as well.

Soon as I get the list compiled, I will post it here.

 

[JohnnyEnglish]

James,

I would certainly also be interested in seeing the more obscure references which you may have found.

As Ed said, the goal is to try and confirm techniques for predicting fatigue in riveted structures. Do you have any experience with FE analysis and random vibration spectra?

Maybe I could give a specific example of the way in which I go about things and open it up for discussion. I shall construct and post soon (if there's demand).

JE.

 

[Sparweb]

Indeed, JohnnyEnglish, the demand exists...

Have you ever checked out the "Def Stans"?



Steven Fahey, CET
"Simplicate, and add more lightness" - Bill Stout