Fatigue assessment without the fudge 2

[corus]

Some new software has been produced that is supposed to eliminate the problems with fatigue classification in FE models and stop the arguments regarding mesh density :

http://projects.battelle.org/verity/

Has anyone any experience with this in practice and is it worth looking at?

corus

 

[drmh]

Let me just give an idea on the difficulties in weld assessment and the ideas to solve them.

The main problem is that the local geometry of a weld is

complex
not known before manufacturing

Due to the complexity one wants to avoid to model the weld in the finite element model, wherefore one wants to stay with simple shell meshes that give the global stiffness of the structure. If one in interested in the fatigue there is a need to get some ideas on the local behavior.

One "old" idea is to use some kind of nominal stress either taken in a given distance of the weld of to extrapolate back to the weld (hot spot appoach) These values than are compared to test data (SN curves) which depend on weld type and loading condition. Clearly to get the stresses in a defined distance from the weld indicates that this approach is highly dependent on the local meshing.

To overcome the second point developments on using a normalized weld structure (R1MS approach) had been undertaken beginning of the 1990ies. In the meantime this approach is the accepted one by the IIW (International Intitude of welding). It shows a small scatter in the used data.

But in principle this would need to use a fine 3D solid model for the complete structure or the use of some kind of substructuring approach. And this is where the force based approaches of Batelle or (more refined) LMS get into the game. The basic idea is to use the element nodal forces and moments in the weld line (of a shell mesh) to back calculate the forces and moments acting on the weld itself which than can be (automatically) used on a solid sub model of the weld.
This can be done for each time step indidually such that the stresses needed for a structural stress approach can be achieved.

See JSAE 20037043 / SAE 2003-01-2772 for details.

drmh

 

[mcclain]

HookedOnHalo,
One of my favorite references is:
Niemi, Erkki, Stress Determination for Fatigue analysis of Welded Components, Abinton Publishing, Cambridge, 1995.
ISBN 1855732130
Regards

 

[marine59]

The submodel approach is a good one, and solves the problem of meshing a large frame with grain-of-sand elements.
HOWEVER: if you build a singularity, infinity will come, followed immediately by crack initiation. As pointed out above, this is not a mesh size problem.

In the 60's it was common to post-grind welds to
1) remove the thermal shrink stress cracks at the toe of the weld, and 2) create a radiused (non-singular) shape which could then be addressed with Peterson's stress concentration factors, or now by sub-modeling.

As for processing real scattered fatigue data into straight lines, it sounds like snake-oil. At least one engine manufacturer envelopes the minimum failure stress data and shifts it to the left by a factor.
Scattered data is, and the only thing you can do to get MINIMUM life is to ignore the optimistic data points.

 

[drmh]

marine59,

the singularity is an artifical one in the stress field modelling a seam weld connection just by connecting shells.
That's in fact the reason to look in the lement nodal forces and moments that have no singularity in the weld line. And yes the singularity is not a mesh size problem.

In the real structure there are continous shapes and there as well are no singularities.

To the scattering issue, again reality is (shoud be) the measure. The failure of a structure and especially a weld is determined by many factors, such that from the point of view of an experiment it is a random process. If you know more about the scattering the better your predictions can be.

So you would not just ignore the optimistic points because they are as important as the pessimistic ones for the estimation of the probability distribution (scattering) of the process. Having collected enough good data you will not geta minimum life (the only strict minimum there is 0) but a reliability of your structure which is much more useful.

Regards