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

 

[xerf]

corus,

that sounds interesting. I wonder what is the mysterious numerical approach behind the software. Maybe a lattice/particle method ???

however, I think the difficulty when modeling fatigue comes also from the variety of phenomena which concur to the final failure

 

[TGS4]

This has previously been known as the "structural stress" method. For FEA, it uses forces as opposed to stresses, lessening the reliance on a particularly fine mesh. The method is based on thousands of cross-industry fatigue tests, and the "mysterious numerical approach" is actually a method for reducing the scatter in the fatigue data. I have seen the data presented by Battelle, and it is impressive how the data has been collapsed to a rather thin line with small ±.

 

[xerf]

Therefore, the mysterious numerical method is still FE method.

For FEA, it uses forces as opposed to stresses, lessening the reliance on a particularly fine mesh.

This is not very clear to me: do you mean that the stresses are not intregrated over the element , or the program does not use the concept of stress?

Also:
I was considering that experimental testing is more prone to producing scattered data than the deterministic FE methods.
Other fatigue programs use libraries of experimental data and known analytical solutions, for example:

http://www.swri.edu/4org/d18/mateng/matint/NASGRO/default.htm

 

[TGS4]

The method is based on nodal forces.

I agree that experimental fatigue data has a large amount of scatter, but the methodology that Battelle uses collapses the data to a very narrow scatter band.

There should be a WRC bulletin coming out on this shortly...

 

[corus]

The technical papers appear to be worth looking at, but why is this only related to welds and not unwelded structures such as within the material or at the edges - away from welds, etc.

corus

 

[xerf]

I took a quick look at their latest paper:

"A structural stress definition and numerical implementation for fatigue analysis of welded joints."
by P. Dong
published in International Journal of Fatigue 23 (2001) p.865–876

It seems that they use a post-processing decomposition of stress field into structural stress components: membrane and bending which have simpler distributions in the cases they present. (Why am I saying "they"??)

However, a relevant quote from this paper:

"It should be noted that in typical finite element based stress analysis, the stress values within some distance from the weld toe can change significantly as the finite element mesh design changes (e.g., [9]), referred to as mesh-size sensitivity in this paper."


No further comment.

 

[prost]

Shouldn't "mesh size sensitivity" be eliminated by obtaining numerically converged FE solutions?

 

[johnhors]

Prost - yes it should !

 

[corus]

prost,
No you can't eliminiate mesh sensitivity. If you're modelling a weld where two planes intersect at right angles then you won't get mesh convergence to a solution at the intersection.

In design standards for fatigue they normally refer to the nominal stress away from the weld toe. Sometimes this is is obtained by extrapolating results up to the position where the weld toe would occur, and thus eliminating any singularity that might occur from the geometry, picked up a high mesh density at that point. I presume this method relies on a reasonable mesh that can elminate the variation that may occur in results due to the over-zealous or bone idle modeller.

corus

 

[prost]

If two planes are intersecting at right angles, if I get your description correctly, this is what some call a 'reentrant corner'--this is a numerical singularity; therefore the stresses will never converge because the exact solution is infinite. No amount of fudge factoring will change the nature of the exact solution.

 

[prost]

I am curious why extrapolating to the weld toe is even necessary? Is there a bi material interface (that is, is the material of the weld modeled as some other material different from the plates you are welding?)? Are the materials elasto-plastic? Or some other nonlinear material? I will have to do some more research on modeling of welds with FEA to understand what is really going on. The Verity method at Battelle appears to me an empirical correlation (that is, based on test results) applied to FEA results to smooth the FEA results which will show infinite stresses if there is a geometric or material singularity.

 

[corus]

prost - In a non-uniform stress distribution it's difficult to ascertain the nominal stress needed for fatigue assessment. One way around the problem is to plot the stress distribution up to where the weld is, and then apply some linear interpolation to the results, or basically just guess, at what the nominal stress is at the toe without the effect of the geometry of the weld. This guessing, or fudging, is something I hoped that this method would reomve.

corus

 

[prost]

I understand then that there are two problems here--1) calculation of fatigue life in a non uniform stress field--such a calculation normally requires the estimation of some stress parameter such as von Mises stress, and 2) calculation of the required 'stress parameter' at the weld toe.

Since the first problem is company specific (that is, there are many ways to estimate fatigue life of a structure--crack initiation and crack propagation come to mind, though each method has many possible variations and fudge factors associated with it), I will address the 2nd problem only.
Since this appears to be the problem of a singularity at the weld toe, doesn't 'extrapolating back to the weld toe' give you the same result as 'extrapolating back to the crack tip'? In other words, garbage. The only question is the singularity's strength. A crack tip is a numerical singularity--the exact solution is infinite stress at the crack tip. Within Linear Elasticity theory, Williams described the stress field as one proportional to a constant, the Stress Intensity Factor KI and the inverse of the square root of the distance relative to the crack tip--at the crack tip, the stress goes to infinity. Since Therefore, any calculation (with FEM, BEM, etc.) of the 'stress' at the crack tip will be meaningless.

It is possible though with experimental correlation to make some average stress computation with linear elasticity models in FEM/BEM in the neighborhood singularity. Sure would love to see how someone does that, though! Perhaps this Battelle software does just that.

 

[corus]

No, extrapolating back to the weld toe isn't the same as extrapolating back to a crack tip of infinite stress. You would have had to have had a crack-tip model to produce such garbage using a totally inappropriate method for estimating the nominal stress.
Fatigue at a weld is based upon empirical data of test pieces that have been subjected to a simple stress field. The problem with life is that nothing is simple. Unless you're modelling a relatively simple structure with easily identifiable nominal stresses that can be related back to these empirical results then you have problems in assessing the fatigue life. Extrapolating to the weld toe, whilst removing the gemoetric stress concentration factor of the weld, gives a conservative estimate of the nominal stress from which to compare with this empirical data.

corus

 

[prost]

Here's my reasoning:

corus, what is the nature of the stress at the weld toe? Is it singular; that is, the exact solution is infinity? There are many singularities--reentrant corners, bimaterial interfaces, and cracks. Each has a different singularity strength. The weld toe looks from my vantage point to be a reentrant corner. If so, then the exact solution for the stress in infinity, therefore it makes no sense to take stresses away from the corner and extrapolate back to it.

If I have made a bad assumption regarding the nature of the singularity at the weld toe, go ahead and set me straight. But if the exact solution of the stress at the weld toe is infinity, then the extrapolation technique is garbage.

 

[corus]

Another way of looking at the stress at a weld toe is to consider it as a peak stress, as defined in many design standards. The practice in these standards is to seperate the peak stress that occurs at stress concentrations (or non-linear thermally induced stresses) from the primary stresses by using linearisation methods. These primary stresses can be thought of as being the nominal stresses at the weld. You could consider the extrapolation to the weld toe as a kind of linearisation method to remove the peak stress caused by the singularity. It'd be difficult to describe the method as garbage as it is a design standard method, and presumably proven to work.

corus

 

[GregLocock]

I'd argue forward from the real-world observation that welds work, rather than saying here is a linearised approximisation to a weld and Omigod it fails.

2000 years of engineering has developed some good solutions, and joining bits of iron by melting them is a technique that has a great deal of practical experience behind it.




Cheers

Greg Locock

Please see FAQ731-376 for tips on how to make the best use of Eng-Tips.

 

[corus]

I'm not sure what Greg is referring to, but to avoid confusion, the linearisation refers to the stress distribution up to the point at which the weld toe would occur, and not the representation of a weld by a straight line.

corus

 

[HookedOnHalo]

Do you have any references that explain method of extrapolating the stress at the weld to avoid the effect of the singularity? I've seen some discussion in a DNV standard that refers to a hot-spot technique but its based on thin material, that is, the weld stress is extrapolated based a distance of 0.5xtk and 1.5xtk of the plate.