What Residual Stress Does After Everyone Goes Home
Before you release anything
A restrained weldment holds real elastic energy and it moves when the restraint comes off. Support the assembly, stand clear of the direction the free end will travel, and release clamps and strongbacks one at a time. Cutting a restrained weldment can close the kerf onto the torch or blade and swing the offcut, so support the offcut and stay out of the plane it will move in. Where the assembly is painted or galvanised, cutting or grinding it is an inhalation route whose constituents come from the coating rather than from the smoke's appearance, needing capture at the tool or a respirator issued under a written program meeting 29 CFR 1910.134. Needle scaling and other impact tools put the work inside a hearing conservation program under 29 CFR 1910.95. Fire prevention and the fire watch for any cutting belong to the hot work permit under 29 CFR 1910.252(a), or 29 CFR 1926.352 in construction.
Why this matters
A weld cools by shrinking, and the metal around it does not let it. What the shrinkage cannot do as movement, it does as force, and that force stays in the part after the arc is off and after the crew has gone home. No gauge on your truck reads it, and it sits at the joint at a magnitude most people would not accept if you called it an applied load. Two things then happen in the field: a part that measured square on the bench comes back out of tolerance after somebody drills, mills or flame-cuts it, and a joint that carries its service load fine cracks anyway, because the service load was never the whole load. This card is about the second half of the contraction, the half nobody measured.
What a weld locks in
Molten weld metal solidifies hot and contracts as it cools. Cold parent metal on either side does not contract with it. The weld ends up stretched, which is tension, and the metal further out is pushed the other way, which is compression. The two balance, because nothing is pushing on the part from outside. That produces a self-equilibrating field: tension in and beside the weld, compression a short distance away, summing to zero across any section you cut.
The magnitude worth carrying is that longitudinal residual stress along the weld commonly reaches the yield strength of whichever metal yields first, the weld metal or the base metal. That figure carries its conditions: longitudinal, at or very near the weld centerline, in a weld on plate free to contract across the joint. Transverse residual stress in the same weld is lower, unless the joint was restrained across its width too, in which case the transverse component climbs toward yield as well. A number without its direction and its restraint condition is not a residual stress figure.
Two consequences follow. The part is already near yield at the toe before you apply anything, so service stress lands on top of a field your calculation never included. And because the field is self-balanced, removing metal anywhere unbalances it and the part moves to find a new balance.
The gate: restraint decides whether you get movement or stress
Restraint is the single gate. The contraction is a fixed quantity set by the weld's volume, the heat input and the material's expansion behavior. Where it goes depends on whether the part can move. Low restraint (a small part tacked lightly on a bench, nothing clamped) gives visible distortion and comparatively modest locked-in stress. High restraint (a member welded into a completed frame, a part bolted to a heavy fixture, a thick section welded to a thicker one) gives very little distortion and a field near yield.
You do not get to avoid both. A fixture does not delete the contraction, it converts it. Distortion prediction and the sequencing that manages it belong to the sibling cards; this card owns the other side of the gate, which is what you are storing when you win on shape.
Two parts, one gate, opposite outcomes
Take one detail welded two ways, and follow the same numbers through both. All dimensions below are illustrative, chosen so the arithmetic is checkable, not lifted from a job.
The part. A 1/4 in mild steel gusset, welded to a base plate with two 6 in fillets on one side only. Squareness is checked with a machinist square at 8 in up the gusset from the corner.
Case A, free on the bench. No fixture, tacked at the ends. After cooling, the gusset has pulled toward the welded side. The square reads an offset of 0.42 in at 8 in, which is about 3 degrees out. It was corrected in the press at fit-up of the next assembly: 0.4 hours.
Case B, clamped in a fixture. Same detail, same welds, same welder, clamped to a heavy bed. Out of cold, the square reads 0.03 in at 8 in, about 0.2 degrees. It measures excellent, it ships to the machine shop, and a lightening hole is flame-cut through the base plate 2 in from the weld toe. The part springs on the table. Re-checked, the square now reads 0.22 in at 8 in.
What the two cases say together. Case B did not avoid the contraction, it stored it. Cutting released 0.19 in of movement, taking the reading from 0.03 in to 0.22 in, against the 0.42 in Case A gave up freely on the bench, which is 45 percent of it, and the balance is still in there waiting for the next cut. Case A cost 0.4 hours of press work at a stage where the part was still loose. Case B cost 3.2 hours to re-fixture and re-machine after it moved, an 8.0x multiple on the same defect, and the movement arrived after the part had already passed a dimensional check.
The call. Neither case is a mistake. Fixturing is correct when the tolerance is on the shape, and it becomes wrong the moment somebody downstream removes metal without knowing the part is loaded. What flips the recommendation is what happens next to the part: if it is going to a mill, a drill, a flame cutter or a hot-dip galvanizing kettle, the fixture bought you a measurement that is only true until the next operation, and the assembly needs either a relief step specified by the engineer or a machining allowance that assumes movement.
Doing the cut. Cutting a restrained weldment releases stored energy: the kerf can close and pinch the torch or blade, and the offcut can swing. Support the offcut so it cannot drop or rotate into you, keep your body out of the plane the part will move in, and cut in the sequence the engineer gives you where the assembly is thick or heavily restrained. If the weldment is painted or galvanized, cutting it releases metal fume that the smoke's appearance will not tell you about, so establish the constituents and the control before the torch lights (see the fume card in References).
What releases it later
The field stays put until something disturbs the balance. Four things reliably do:
Metal removal. Machining, drilling, flame cutting or gouging removes part of the balancing compression or tension, and the remainder redistributes. This is the classic surprise: the flatter and squarer the part measured, the more it had stored, so the parts that pass a dimensional check best are the ones most likely to move on the mill.
Heat. Local heating lowers the yield strength where it is applied, letting the stored stress relax there and reappear somewhere else. That is the principle flame straightening runs on, and it is also why a second weld near the first moves a part that was stable.
Service loading. Applied stress adds to residual stress at the toe, and where the sum reaches yield the material yields locally and sheds some of the field. That is why some weldments settle after their first load cycles, and it is not free: it is a plastic event at a notch.
Time plus hydrogen plus a hard microstructure. Residual stress is one of the three legs delayed hydrogen cracking stands on, alongside dissolved hydrogen and a crack-susceptible microstructure. The mechanism belongs to the hydrogen card; the leg this card owns is the one you built in the moment you restrained the joint.
What actually lowers it, and what only appears to
| Approach | What it actually does | Who owns the numbers |
|---|---|---|
| Post-weld heat treatment | Holds the part hot enough that yield strength falls and the stored stress relaxes by creep, then cools it slowly and uniformly | The fabrication code in the edition your contract adopts, the material specification, or the engineer of record |
| Reduced restraint and a planned sequence | Lets contraction happen as movement you predicted rather than as stress you did not | The sequencing card, and the WPS where it specifies sequence |
| Lower heat input, smaller weld | Less molten volume contracting, so a smaller field, but never zero | The WPS, within its qualified range |
| Peening intermediate layers | Locally works the surface and can redistribute stress, but it is restricted by fabrication codes and prohibited on root and final layers in common ones | The adopted code edition, explicitly, before you do it |
| Straightening the part cold | Changes the shape, not the field, and adds its own worked-metal stress | Nobody, and that is the point |
| Letting it sit | Nothing at room temperature on any schedule you care about | Nobody |
Two of those need a sentence more. Post-weld heat treatment is not yours to decide: soak temperature, heating and cooling rates and hold time are set by the material and the code, and a wrong cycle can temper the part out of its specified properties. Peening is the one people reach for from memory, so read the adopted edition's wording before an operator takes a needle scaler to a weld, because "we always did it" is not an adoption path. Needle scaling is loud and throws particles, so hearing protection under a program meeting 29 CFR 1910.95 and full eye and face protection go on before the tool runs.
Convention and basis, so the number travels
If a drawing, a report or an engineer hands you a residual stress figure, three things travel with it or it means nothing. Direction: longitudinal along the weld, transverse across it, or through thickness, which are different magnitudes in the same joint. Location and depth: surface at the toe, weld centerline, or a stated depth, and a surface value is not a through-thickness value. Method and character: hole drilling gives a near-surface value with a measured spread, X-ray diffraction a value in a shallow surface layer, and both are point estimates with their own uncertainty rather than a bound on the joint. A single point does not characterize a field.
How to verify you got this right
Print the figure next to the criterion, every time:
- Did the fixture convert movement into stress? Case A moved 0.42 in at 8 in; Case B moved 0.03 in at 8 in. Difference 0.39 in, which is the movement the fixture converted into stress. Of that, 0.19 in reappeared on the first cut, taking Case B from 0.03 in to 0.22 in, which is 49 percent of the converted amount. The conversion is real and it is measured, not asserted.
- Does the correction close? 3.2 hours downstream against 0.4 hours at fit-up is 8.0x. Both figures are unbilled correction hours, so the ratio compares like with like.
- Is any rounding running in the flattering direction? 0.19 of 0.39 is 48.7 percent, reported as 49 percent. Rounding it up does not flatter the fixture, because a larger reported release is a larger admission that the shape was bought with stored stress.
- Does the example obey the sibling rules? The flame cut in Case B is hot work, so it runs under a permit and a fire watch per the hot work SOP rather than under this card. The cut edge left by that torch is not a weld-ready edge, and if anything is welded to it, the thermal cutting card governs the removal before the arc, not this one.
- Did I claim a stress magnitude without its condition? The yield-magnitude claim above is stated as longitudinal, at the weld centerline, with the transverse case called out separately. If it appears anywhere without those words, it is wrong.
References
- 29 CFR 1910 Subpart Q (welding, cutting and brazing, general industry) and 29 CFR 1926 Subpart J (construction), with 1910.252 and 1926.352 as the hot work counterparts for the cutting steps described here.
- 29 CFR 1910.95, occupational noise exposure, for the hearing conservation program covering needle scaling and other impact tools.
- AWS D1.1 Structural Welding Code - Steel, in the edition adopted by your contract or by the authority having jurisdiction, which owns post-weld heat treatment parameters and the restrictions on peening. It binds the fabricator through that adoption, not on its own.
- See related: What Distortion Is and Why It Is Predictable; How to Sequence Welds to Control Distortion; Why Hydrogen Is the Enemy in a Weld; What Thermal Cutting Does to the Edge You Are About to Weld