Modelling Partial Penetration Welds in FEA

[Speedy]

Problem as follows;

We have a large cast component which is the main part of a structural frame of a hydraulic press type machine. In service loads are high cyclic tensile, clamp unclamp etc.

To save on cost & lead time it is advised that we produce the part from welded plates. Stress relief, weld inspection (ultrasonic)& process control are all advised.

My main concern with the part is the level of welding. The majority of welds are partial penetration welds. Most plates are about 50mm thick. We feel that this will greatly weaken the structure as compared to our existing cast base.

Such welds we feel should also show up as elevated stresses on a completed FEA as the bonded metal to metal contact area is greatly reduced. In the FEA, Welding was simulated by merging the nodes along the entire interface surface of the plates.

Does this accurately reflect the proposed level of partial welding?
Should the nodes be merged only along the welded interface and not the entire interface area as seems to be the case.

I have raised this issue with the FEA guys but they say it is fine. Surely when you reduce the bonded area at the plate interfaces there is a large reduction in strength.

Appreciate any comments.

Speedy.

 

[johnhors]

Speedy,

FEA is not an exact art, it is a tool with limitations that at best will give you an approximate answer (if all best practices are followed). What you are asking, to have only partial merging of the nodes at the interface falls outside the limits of acceptability for a global model. Instead a highly refined sub-model of the weld area would be required, even then I would approach the results with caution.

Because by introducing an internal surface of unconnected nodes like this, you are effectively introducing a crack or split in the mesh, this then creates a singularity at each end of the split. The global FE model cannot cope with this, the model stiffness matrix would become ill-conditioned and the subseqent results highly suspect.

Imagine a plate welded to another at 90 degrees then for bending one weld line would see tension whilst the other sees compression. Thus nodes joined in the middle (as your FEA guys have done) are more less on and close to the neutral axis, so these nodes should transfer less load. On the real thing the unwelded mating surfaces are effectively clamped together by the welds and will tend to deflect as one, transmitting forces by surface contact, whilst if you model does not have these internal nodes merged then each face is free to penetrate the other (which could well happen!), which is obviously incorrect.

I agree with the approach of your FEA guys as being "fine". Also I think your assumption on there being "a large reduction in strength" is misguided.

 

[corus]

I don't agree with Johnhors as you can't guarantee that the unwelded nodes will be in contact and so able to transfer the loads. A slight gap will leave a crack as he says, whether it be in compression or in tension. You will see highly concentrated stresses at the crack tip but these will be localised and not affect the global results. Basically you have bad welding practice and you should include a run on the opposite side to the partial penetration weld.

corus

 

[crisb]

From your description, it appears that you are worried that the welded component will have a much shorter fatigue life than the cast component. You therefore need to carry out a fatigue analysis on the welds. I dont think its possible to generalise until you look at the actual loads in the welds. This is probably more important than worrying about the intricacies of the FEA.
I would trust your gut feel until you have proof that it wrong!

 

[johnhors]

Corus, in FEA it is not safe to assume that models which incorporate a singularity in the mesh will produce dependable results elsewhere. All solvers I have ever used check pivoting ratios in the stiffness matrix, depending on the numerical severity, they at the minimum issue a warning message or else stop the analysis procedure when the ratio goes beyond some pre-determined limit. Excessively large pivot ratios will without question corrupt the whole analysis. This is not a question about machine precision as this scenario should always be avoided.

The OP thought he could improve accuracy by introducing a crack in the mesh, whilst in reality all he was doing was invalidating the model.

 

[GBor]

Speedy,

There are many issues with this type of analysis as the others have raised, but it sounds as though your FEA guys may need to learn a little bit more about their own software as well as FEA in general. What element types are they using? Because it sounds as though they could be using plate elements, but unless your structure is very large, 50mm plates may need to be modeled as brick elements.

Also, common sense should tell you that the reduced cross-section of a partial penetration weld will likely reduce the strength of your structure. Add weld porosity, heat affected zone...there are many reasons for you to be concerned. Perhaps the FEA people aren't concerned because your margin of safety is extremely high?

 

[corus]

Every model that has a 90 degree juncture or any sharp change in geometry will have, in theory, a singularity. This doesn't invalidate every model, nor does it produce invalid results elsewhere in the model. The same applies to a model which will have a crack in the geometry. I think it's Saint Venant's principle of localised stresses not affecting the global distribution. I've ran lots of models with such defects and have had the results validated. The results around the crack will show a high peak of course, and won't be of much use.

corus

 

[johnhors]

A crack introduces a zero degree corner into a mesh not 90 degrees. The effect is not the same.

 

[Speedy]

Guys,

Thanks for all your informed opinions.

To be honest, I have very little knowledge about FEA, only what I vaguely remember from college.

My point though, is as follows.

I appreciate what Johnhors is saying about bending on a 90deg welded plate. The larger stresses are at the area of the area of the welds.

Take for example though an "I" Beam in 3 pt loading. If the beam is welded (part penetration) of 3 flat plates, the transverse shear at the beam ends and at the plate interfaces will be greatly increased by the reduced contact area.
Our structure is loaded similar to 3 pt loading so similar transverse shear at the plate interfaces. The transverse shear wont be transferred by the non-welded metal to metal contact area.