Welding distortion

Why does a welder deliberately set two plates out of line before joining them straight?

A fabricator wants two plates joined flat. So they set them up deliberately not flat — tilted apart by a few degrees, or bowed the wrong way — clamp them there, weld, and the assembly pulls itself straight as it cools.

Setting a job up wrong to get it right looks like superstition, or at best a knack. It is neither. It is a calculation, and it rests on the fact that a weld is not a joining operation with a thermal side effect. It is a thermal operation that happens to leave a joint.

Start with what the arc actually does. It melts a small volume of metal and heats a larger volume around it, while the rest of the plate stays near room temperature. That is a steep temperature gradient in a material that expands when heated and contracts when cooled.

Now follow the sequence carefully, because the asymmetry is easy to miss.

Heating. The hot zone tries to expand. It cannot, because it is surrounded by cold metal that is not expanding. So instead of growing, it is squeezed — put into compression by its own surroundings. And here is the crucial step: hot metal is weak. Rather than pushing the cold plate aside, the hot zone yields, upsetting plastically and becoming permanently shorter and thicker than it would have been.

Cooling. Now the same zone contracts. But it is contracting from a state that has already been plastically shortened, and as it cools it regains its strength. So it pulls — hard — on the surrounding cold metal, and this time it does not yield, because it is no longer hot enough. The weld zone ends up permanently shorter than the metal around it, in tension, with the plate pulled toward it.

That single asymmetry — yielding while hot, pulling while strong — is the whole mechanism. The weld does not merely shrink; it shrinks irreversibly, because the expansion was absorbed by plastic flow and the contraction is not.

Now the direction of the distortion follows from where the shrinkage sits. Shrinkage along the weld line pulls the plate lengthwise. Shrinkage across it draws the two plates together. And most visibly, a weld that is wider at the top of the joint than the bottom — which any single-sided V-preparation is — shrinks more at the top than the bottom, so the plates rotate about the joint and the assembly folds up. That is angular distortion, and it is the one pre-setting is usually aimed at.

So the counter-measure writes itself. The distortion is predictable in direction and roughly repeatable in size for a given joint, plate thickness and procedure. If you know the assembly will fold by three degrees, set it three degrees open and let the fold bring it to zero. The welder is not preventing the distortion. They are spending it.

That framing also explains the alternatives, which are all versions of the same idea. Welding alternate sides of a joint balances the shrinkage about the neutral axis. Back-step and skip sequences spread the heat so no one region dominates. Balanced fillets on both sides of a web cancel. Heavy clamping restrains movement — but note what it does not do: it converts distortion into locked-in residual stress, which is a different problem rather than a solution, and can bite later when the part is machined and the stress released.

The metal does not shrink because it "used to be liquid". It is tempting to attribute everything to solidification shrinkage of the deposited bead, and that contribution is real but small. The dominant effect comes from the much larger heat-affected volume that never melted at all — it expanded, yielded while hot, and contracted while strong. This matters because it predicts that distortion scales with heat input rather than with the amount of filler deposited, which is why a low-heat process on the same joint distorts less even though it deposits similar metal.

Pre-setting is calibrated by experience, not by calculation, and that is a real limit. The direction of distortion is reliably predictable; the magnitude depends on plate thickness, joint preparation, restraint, sequence, process, travel speed and the fit-up on the day. Shops build up a feel for their own repeated jobs, and pre-setting works well for production runs of the same part. For a one-off, it is a much weaker tool, and the honest practice is to favour sequencing and balanced welding over trying to guess a number.

Clamping and pre-setting are not interchangeable. Restraining a job holds it in shape while locking the stress inside, so the part is straight but stressed; pre-setting lets it move and arrive straight with less locked in. Which is correct depends on whether the part will later be machined, heat-treated, or loaded in a way that cares about residual stress. Treating clamping as a universal answer is the common error, and it is why parts sometimes distort on the machine after being welded straight.

Some of the shop lore is dubious. Hammering a weld to "relieve stress", and the various claims about welding sequences that supposedly eliminate distortion entirely, are far more confidently stated than the evidence supports. Peening has legitimate uses, but as a general stress-relief method it is contested. Genuine stress relief means a controlled thermal cycle, and that is a specified process rather than a knack.

The falsification test. If the mechanism is plastic upsetting while hot followed by elastic contraction while strong, then reducing heat input at the same deposited volume should reduce distortion, and pre-heating the whole plate — which flattens the gradient — should reduce it further. If distortion tracked the mass of filler metal regardless of heat input and gradient, the account would be wrong and solidification shrinkage would be the main term.

• Why residual stress can reappear as movement when a welded part is machined.
• How balanced and back-step sequences distribute heat to cancel the same effect.
• Why low-heat-input processes distort less at the same deposited volume.