Weld Analysis at Throat of Model 4

[Penguineer]

Hello All,

In my quest to determine the most appropriate method of static weld analysis with FEA I have come across at least a half dozen methods. All of these seem to have their own merits but I can rarely make them agree with each other, or even the hand calculations. To the point: I wanted to try sub-modeling the (fillet) weld joint and splitting the surface through the throat plane. From there I figure I could measure maximum shear stress and compare against distortion energy theory (Sys = 0.58*Sy), which for an E70 weld is 33,640 psi.

My rationale behind this is that a weld always fails in shear and the failure would occur at the throat. Furthermore, the hand calculations seek to determine the resultant stress in the throat and either compare them to an AISC/AWS allowable or compare them to an upper stress limit, depending on whether you're reviewing the weld as a line or as the throat area. I figure the focus of all of this is to see the load passing through the throat of the weld and to weigh that against the appropriate stress limit, so this FEA approach might be reasonable.

Has anyone tried anything similar to this and could comment on the method? I would like to hear your feedback on this or perhaps another method that you prefer.

Thank you.

 

[rickfischer51]

The way I analyze welds with FEA is to extract reactions at the joint and take them into a spreadsheet, where I calculate weld stresses and allowables according to what ever code I am using. The stresses calculated in the codes are not really stresses, they are "stress conventions." The codes were written before FEA and depend on simplified hand calculations, which are somewhat imprecise. They are compared to an allowable that is suitably knocked-down to reflect the impreciseness. Your finite element model will likely result in a different stress than the code, so it may not be appropriate to compare it to the code allowable. So, take the reactions, calculate the stress based on the code you are using, and go from there.

I was originally tempted to do what you proposed. Then I read Shigley, Mechanical Design Analysis, Chapter 9. He starts a rigorous derivation of the stress in the welds, then says (page 417, 4th edition) "However, for design purposes it is customary to base the shear stress on the shroat area and to neglect the normal stress altogether." So the formula he gives is not a real stress, but a stress convention. Also, there is a paper by M. A. Weaver, "Determination of Weld Loads and Throat Requirements Using FEA with Shell Elements: a Comparison with Classical Analysis" that is pretty good. It's easy to find by googling.

Rick Fischer
Principal Engineer
Argonne National Laboratory

 

[Penguineer]

Thanks for the input Rick, I like your approach. I'm familiar with Weaver's paper and had considered getting his addon for ANSYS, but I would like to explore other techniques before requesting the $8,000 and justifying it to management. It seems to me that acquiring reaction loads from a joint in a shell modeled structure is preferable because you're no longer dealing with the mesh-sensitive stress measurements. The difficulty I have with this approach is that I don't have a good feel for how reliable this approach is. If I recall correctly, Weaver's software aligns the local coordinate systems where the measurements are taken with respect to weld normal/transverse/longitudinal directions. I'm not sure what other pre-processing is to be done to obtain the necessary force reactions.

In following this approach, I don't know if I can simply output the nodal reactions, calculate the resultant and compare to AISC/AWS allowables (which I obtained from Blodgett's book). If you have some more information on this approach, preferably in ANSYS, it would be greatly appreciated.

 

[rickfischer51]

There are two methods that can be used. One is to apply the loads to the weld group as a whole (Shigley, Blodgette, etc), while the other is to calculate weld stresses on a node by node basis down the length of the weld (Weaver). The first method is consistent with the methods used in the various codes, so the code allowables can be used. The second method captures stress variation due to local stiffness conditions, but can produce locally high stresses that are difficult to satisfy with the code allowables. This is particularly true at the end of a fillet weld. These locally high stresses get averaged out with the first method, and I believe the allowables were adjusted down to account for this. If you use the second method, it would seem that the allowable should be increased, but by how much? In the end, the first method is the cleanest and easiest to use, but the allure of a more correct solution offered by the second method is still there.

Attached is a simple model that demonstrated how I do a weld analysis. Everything through solution is done in WB, then I open the model in MAPDL and run the macro. The macro creates a file called weld_data.txt that can be opened in Excel, where the calculations can be done. To make this work, the WB model needs to have named selections consisting of edges at each weld. The selections are named weld1, weld2, etc. Next, place a local coordinate systems at the centroid of each weld or weld group, oriented as per Shigley. In the Details menu, set Coordinate System to Manual, and number them so that the csys at weld1 is 101, at weld2 is 102, etc. Then, in Analysis Settings, set Output Controls, Nodal Forces to Yes, and in Analysis Data Management, set Save MAPDL db to Yes. Place the macro in your macro home folder or in the weld demo-files\dp0\sys\mech folder.

Hope this helps. Have fun.

Rick Fischer
Principal Engineer
Argonne National Laboratory

 

[Penguineer]

I ran this sample model in ANSYS, I was actually able to avoid using APDL and keep it in workbench by integrating the macro code as a Command instead. All works great, but fundamentally I have some questions.

After extracting the loads and moments at the joint, the section modulus for the joint is to be determined from weld properties treated as lines. I assume that if we're dealing with a single sided weld, the weld is treated as being at that joint (the line of intersection) rather than offset, so that Sw = d^2/6. For the case where we have a fillet weld on both sides, does the same theory apply, with the exception of using Sw = d^2/3 instead?

I attached a couple of slides for calculating the weld stress at the worst point in weld 1. In this case the moment about the local coordinate system Mz = 0. If this were non-zero, then this would result in a torsional stress in the shear plane (XY plane). Wouldn't this torsional stress sum with the transverse shear at point A?

Lastly, there is a relatively large moment about the Y-axis (My) that is unaccounted for in these calculations. Referencing my literature I cannot find cases where a plate is bent about the weld axis and the moment is accounted for in the calculations. It seems only the shear (Fx) from the force producing the bending is accounted for. This bending of course results in the most extreme stress, but since the section modulus about Y-Y is zero, a bending stress calculation cannot be solved for. How does this bending moment (My) get resolved and combined with the other stresses? I can solve for bending stress from My if I use throat area calculations, but then that gives me a direct stress and not a load per unit length value.

Thanks again for all this, you've been more than helpful.

 

[rickfischer51]

1. Not sure of your nomenclature, but yes, I think. See the top two entries in the attached table from Mechanical Engineering Design by Shigley (I have the 4th edition).

2. Yes. What I do is calculate each shear separately, then add them vectorally. Shigley suggests using Mohr's circle to find the maximum shear stress, but I believe the practice is to treat all stresses as shears and add them.

3. That has always bothered me. I think the assumption has been the the plate is thin so the effect of the moment is small. I go with what's in the tables because I think that is factored in to the allowables. You could treat it as a weld group like in the third entry in the table.

Rick Fischer
Principal Engineer
Argonne National Laboratory