Weld-toe stress in a shell model: where it is read and what it misses

A shell element stands for the wall by its mid-surface. Where a nozzle meets a shell, the two mid-surfaces cross on a curve, and the mesh joins the parts there by sharing nodes. On the shell's side that curve is half a nozzle wall inside the nozzle; on the nozzle's side it is half a shell wall inside the shell. A fillet weld adds its leg on top. A stress read on that line is not a stress on any surface of the real part.

The shell model also has no weld in it. The joint has no fillet, no added stiffness and no notch. Whatever the model says at the junction is the answer for two plates meeting at a sharp corner on their mid-surfaces.

The physical place to check is the weld toe, where the weld meets the plate surface. The model places a probe beside every node of the junction line, at the toe distance along each side, and reads the stress there from the element it falls in. The weld legs come from the job's inputs. The strip between the line and the toe is the weld zone: it is drawn on the contour plot, but no check is made in it.

In our checks against an ASME PTB-3 example and an axisymmetric solid model, PL read at the toe agreed to between −7 % and +3 %. Read on the intersection line, the same stress was 32 % to 67 % high.

The toe is right for the primary checks, but it has a weakness for the secondary one. With no weld stiffness, the shell model under-reads the structural bending at the toe: in our studies by 13 % on one PTB-3 stress classification line. The value at the intersection line, linearised from the forces and moments that cross it, does not have that problem. It is set by equilibrium, it does not depend on the mesh, and in our studies it came within about 3 % of a solid model's value at the toe.

So the model splits the work. Pm, PL, PL+Pb and the local-failure check are read at the toes. The P+Q range takes the larger of the toe value and the line value. The report states the toe distances it used on each side.

Peak stress is the part that comes from the notch itself: the weld profile, the toe angle, the local shape. Division 2 uses it only for fatigue. A shell model cannot produce it, because the notch is not in the model. There are two ways to get closer to it.

• A solid model with the weld in it. The workbench has an optional solid detail of the junction: the fillets meshed as 45° triangles, the block cut 2.5√(R·T) beyond the weld and driven by the shell solution, with stress classification lines through the wall.
• A fatigue method built for welds. The structural-stress method of Division 2 works from the mesh-insensitive membrane and bending at the weld line, which is the quantity a shell model does give.

When a job carries load cycles with a count, the workbench runs Division 2 fatigue screening and then fatigue analysis along each weld line, on both faces, by the structural-stress method and by the elastic method with a notch factor. Where a P+Q range exceeds SPS, ratcheting is checked by the Bree diagram. These checks are built and in early access. How fatigue and Bree are offered on the application cards is still being decided, so they are not described here as a finished feature.

Whatever the method, a model of one attachment and a piece of shell answers for that attachment's welds only. It does not see cycling at other nozzles, supports or closures, or cyclic loads that never reach the attachment. Joint misalignment and weld peaking are not modelled either.