Fatigue - Part Life Prediction

[udme98]

What is a good basis for determining if a part will fail due to fatigue damage? Is it valid to take the fatigue strength of the material divided by three (3 sigma approach) and use this number as the limit?

 

[IRstuff]

Does your load vary that much? Does the material vary that much?

If not, wouldn't you be wasting material and over-designing?

TTFN

FAQ731-376

 

[udme98]

I hope not, I am new to MIL-STD-810F and that is what I was using as a basis and was told by my fellow co-workers to take the fatigue strength and divide by three and make sure the stress is below this value.
I have the FEA results and I am trying to decide if the part will function under the vibration loading.
What is the best way to determine the useable life of a part?

 

[IRstuff]

There's lots of stuff on the web:

http://en.wikipedia.org/wiki/Fatigue_(material)

http://www.statisticalengineering.com/random_fatigue_limit.htm

http://www.key-to-steel.com/Articles/Art142.htm

TTFN

FAQ731-376

 

[BobM3]

There is software that will look at time histories of the strain at a point and, given the material properties, will predict life. Problem with that is one severe overload will make the prediction invalid. However, if you are sure of the loads it would probably give you the best life estimate.

 

[ivymike]

I must not understand something here... is the "three sigma" approach really to divide the material strength by three, or is it to use a material strength which reflects the nominal minus three standard deviations (since real materials vary in strength)?

 

[electricpete]

Mike's way makes a lot more sense to me.

=====================================
Eng-tips forums: The best place on the web for engineering discussions.

 

[IRstuff]

It's a rough rule of thumb to get some cycles out of the S-N curve:

http://en.wikipedia.org/wiki/Fatigue_(material)#The_S-N_curve

A factor of three puts you at over 10^5 cycles.

TTFN

FAQ731-376

 

[cbrn]

Hi,
Life is a stochastically-driven phenomenon, so the "safety factors" with respect to the Richter curves are to be interpreted as "probability not to trigger a damage". I'll try to explain better...
The Richter curves are generally refered to 50% or 90% of survival probability; that means that, when you find 1E5 cycles for Sa = 85 [MPa] with a 1-sigma Richter curve, you have 0.5 probability to reach this number of cycles.
If you want to be much more confident of your life prediction, you have to scale-down the Richter curve by a factor which corresponds to reaching the desired confidence. 3-sigma, or in other terms 99,7% of confidence if I remember well, corresponds to "almost-certainty" and will approximately scale-down a 50%-confidence Richter curve by a factor of 3 in N. For constructional steels, the dispersion factors between 10%-confidence and 90%-confidence S-N curves are approximately 1.5 in S and 4 - 5 in N.
Sincerely I don't know where this "rule-of-thumb" of dividing the resistance by 3 comes from. It sounds a little absurd because, depending on the noth factor, the surface finish factor, the ambient factor, the dimensional factor, the shape factor, and the mean stress Sm (see Goodman-Smith / Haigh...), the scale-down factor between gross-stress allowable for static condition and for fatigue condition can vary WIDELY, from 1 (no fatigue effects, or unrelevant) to 5, 10, ...

Regards

 

[udme98]

Thanks for all the input.
I am using 6061-T6 aluminum for the material which I believe has a fatigue strength of 14ksi.
At first I was taking the 14ksi divided by three (4667 psi) and using this value to compare to the FEA results and trying to modify the design to pass this limit. The longitudinal direction was the worst case situation.
Would using the 3 standard deviation approach yield more realistic results? The part has been redesigned to pass the limit but now looks cumbersome and weighs more than it should. Keep the ideas/suggestions coming. Thanks.

 

[mechengdude]

I believe the technique you are referring to is detailed in Vibration for Electronic Equipment by Dave S. Steinburg. Reference page 250 in the chapter Understanding Random Vibration. The technique is well accepted and conservative. There are some gross assumptions relative to determining the systems resonant frequency.
In reality the load put on the part is determined not simply at the resonant frequency but across the entire vibration spectrum. Without software to calculate modal shapes and induced stress which is usually determined by the mass participation factor, you are stuck with making a hand calculation.