Every weld pulls. Filler arrives liquid, wets both edges, freezes, and then contracts on the way back to room temperature. That contraction does not care whether the workpiece is a titanium canister or a stainless frame. It cares about how much metal was deposited, how hot the surrounding material became, and how firmly the joint was held while it cooled. Treat the movement as a quantity that can be estimated and the job becomes planning. Treat it as bad luck and the job becomes rework.
The reason distortion feels random in most shops is that the result is only seen at the end. A part is tacked, welded, released and checked, and the number that comes back is a single combined figure for a dozen decisions made hours earlier. Break the same part into stages and measure between them, and the movement becomes attributable. This much came from the root, this much from the cap, this much from the fixture letting go.
Planning starts before the first cut. The section is read, the joint count is counted, and the sequence is written down while the part is still a drawing. Material choice changes the technique later on, but the geometry of the problem stays the same whether the alloy is titanium or stainless.
Transverse shrinkage closes the gap across the weld. Longitudinal shrinkage pulls along the bead, and it is what shortens a long seam and bows a flat plate. Angular distortion rotates the two sides of a joint because the top of the weld shrinks more than the root. All three are present in a real joint, and which one dominates depends on joint type, wall thickness and how much restraint the part carries.
A butt joint in thin sheet is mostly transverse and angular. A fillet against a stiffener is mostly angular, because the shrinking bead sits off centre on one plate and drags that plate toward the weld. A long seam on a canister is mostly longitudinal, and that is the one that quietly changes overall length while the diameter still looks acceptable on the bench.
The practical consequence is that one fix cannot chase all three. Clamps address angular movement but convert it into stored stress. Sequence changes address longitudinal movement. Joint preparation addresses how much metal has to be added in the first place, which is the only one of the three that reduces the cause rather than redirecting the effect.
Any welded section has an axis through it where the shrinkage forces balance. Metal above and below that line pulls against each other, and the side with more deposited weld wins. That is why a tee joint curls toward the fillet side, and why a tube seam welded only on one clock position bends along its length.
Once you know which side is winning, the correction is geometric rather than thermal. Depositing a matching bead on the opposite side balances the pull. Moving the seam toward the neutral axis reduces the lever arm. Changing a welded joint into a formed or fastened one removes the shrinkage entirely, and sometimes that is the cheapest redesign available.
On thin-wall titanium the lever arm matters more than the bead size, because the section is flexible and the material has little room to absorb a mismatch. A part that would tolerate an off-centre seam in heavier stainless will visibly bow in titanium with the same deposit. The fix is not to weld cooler and hope. It is to move the seam or balance it.
A closed section such as a canister resists movement far better than an open one such as a bracket. The closed shape distributes the pull around the circumference, so the distortion appears as ovality rather than as a bend. Open sections have nothing to share the load with, so a single bead can rotate a flange by a visible amount.
Long parts accumulate. A short joint that moves a little is tolerable, but the same movement repeated down a seam shows up as a bow at the ends. This is why pipe runs and long silencer bodies are welded from the middle outward or in a stepped pattern instead of straight through from one end.
Restraint is the third variable. A part welded while bolted to a heavy fixture is restrained during cooling and springs when released. A part welded free is distorted immediately but carries less locked-in stress. Neither is automatically correct. The decision belongs to whoever knows whether the final assembly needs dimensional accuracy at the flange or toughness in the seam.
Allowance is simply deciding, in advance, that the part will be built slightly wrong so that it finishes right. A flange face that will pull can be set slightly proud. A tube length that will shorten can be cut slightly long. An angle that will close can be opened before welding. None of this requires prediction to be exact, because a small planned error is easier to correct than an unplanned one.
The allowance has to be recorded, otherwise the next run repeats the experiment. We write the pre-set value on the drawing and on the job card, with the direction of the correction, so a different welder on a different shift produces the same part. The record also tells you when the allowance has stopped working, which usually means a material lot or a fit-up habit has changed.
Straightening is the fallback, not the plan. Press correction works on mild steel with patience. Titanium and thin stainless are less forgiving, and each correction adds cold work where the weld already has residual stress. A part that has been straightened twice is a part with a story, and that story usually ends in a fatigue crack at the toe.
A fixture holds a part in the shape you want while it is hot, which is useful. What it cannot do is remove the shrinkage. The contraction still happens, it simply happens against the clamp instead of against free air, and the difference is stored as elastic strain. Release the clamps and the part moves toward wherever that strain points.
So a fixture buys dimensional control during welding and pays for it at release. That trade is worth making when the part is then machined or when the assembly is welded in place. It is a poor trade when the part must be dimensionally stable as a free body, which is the usual case for an exhaust component that has to bolt to a vehicle without preload.
The misleading part is the inspection. A part measured in the fixture always passes. The number worth recording is the one taken after release and after the part has cooled fully, because that is the geometry the customer receives.
Four factories and a plant of about 53,819 sq ft give us enough different jobs to see the pattern repeat. The parts that arrive back for correction are almost never the ones with a difficult joint. They are the ones where the sequence was decided at the torch rather than at the drawing, and where nobody measured between stages.
Titanium fabrication is the clearest teacher. The material is stiff enough to hold a bad shape and light enough that a small imbalance shows. A one-sided seam on a thin section will pull the part visibly, and no amount of careful arc control compensates for a joint placed off the neutral axis.
The habit that survives is boring. Mark the datum. Weld in a written order. Measure after each stage. Compare with the allowance. Adjust the allowance rather than the part. That loop is what keeps a fabrication shop out of the straightening press.
If you are specifying a welded fabrication, the questions that decide whether it arrives flat are mostly geometric. They are worth settling before a mould or a fixture is built, because both are expensive to change afterwards and neither can fix a design that asks the weld to hold a shape the section cannot hold.
You can see the range of work we build on our titanium fabrication components pages, including plate and engine-bay parts, and compare material behaviour across the shop by material section. Fabricated tube and pipe work sits under exhaust pipe, while full systems are filed under stainless steel exhaust systems and TIG titanium welding parts. The plant behind that work is described on our about us page.
