Accounting for stress concentration in fatigue analysis

[Steev002]

Hey everybody,

I have a question that may seem obvious but I can't find any info that can answer it...

Say I run a FEA simulation on a notched specimen with a stress concentration of Kt=3. If I want to see if the specimen will survive an "infinite" number of cycles, do I check my actual FEA stress against an SN curve for an unnotched (Kt=1) specimen OR an SN curve for a notched specimen (Kt=3) if the data is available? MMPDS provides SN curves for different Kt values.

When using a classical approach (assuming Kt=Kf), either the endurance limit is factored by Kt or the actual stress found using hand calculations is factored by Kt.
Because the FEA model already accounts for the stress concentration, I would think that I should check the actual stress against the SN curve for an unnotched specimen (Kt=1). Is this logic correct?

Are the notched SN curves only useful for stresses found using hand calculations, as they account for notch sensitivity?

Thanks

 

[Mustaine3]

I'm not an expert on that topic, but I think you could use K-factor or J-integral (coming from FEA) with the paris law for a model with crack. Or for an uncracked model the classical fatigue approach, like it is used in fatigue codes like fe-safe or femfat.

 

[rb1957]

it is best to use an SN curve that matches your Kt, if you've got a good handle on the far field stress.

I am doing something similar right now. I have an FEA of a fitting with a stress concentration, and I have the peak principal nodal stress; it's hard to envision a far field stress. I have a pretty small mesh size so I'm satisfied with the accuracy of the model. So I plan to use Kt = 3 SN data, reading in at 1/3 of the FEA stress. Now I'm lucky in that even if the stress was 40% greater than the FEA prediction, I still have an acceptable safe life, with a very conservative safe-life factor (10, 'cause we won't do a fatigue test of these fittings).

another day in paradise, or is paradise one day closer ?

 

[ESPcomposites]

You can also evaluate how the endurance limit "scales" as a function of Kt. For example, if you can observe that the material scales well as a function of Kt (based on unnotched and notched S-N data), then you can probably scale to the particular Kt from the FEM. If it does not scale well (function of the material) then you wouldn't want to use the FEM Kt and unnotched data. In that case, you might be better of scaling with respect to the nearest FEM Kt to S-N Kt (or just use a conservative Kt value from the S-N curve). If you could always use the unnotched data and the Kt, there wouldn't be much need to test notched specimens. But there is a material effect, which is why we need notched S-N data.

Brian

http://www.espcomposites.com

 

[realtime2]

Your statement:

"Because the FEA model already accounts for the stress concentration, I would think that I should check the actual stress against the SN curve for an unnotched specimen (Kt=1). Is this logic correct? "

is correct if you mean by "actual stress" the FEA predicted hot spot stress.
Assuming long life fatigue limit region, elastic conditions etc.

As the other contributors have pointed out most situations are not simple
plates with a hole and thus it is best to also plot any real test data of
the notched component if available.

 

[rb1957]

if you believe the nodal stress on the surface of the concentration, then you have captured the peak stress. You "can" use this with Kt=1 SN data but I wouldn't trust the result. I'd prefer to use SN data for a Kt (maybe 3) with a stress of FEA_stress/Kt

another day in paradise, or is paradise one day closer ?