How Undercut and Overlap Each Form

Why this matters

Undercut and overlap are the two opposite ways of getting the weld toe wrong, and the toe is where fatigue cracks in welded structures overwhelmingly start. Both are surface defects, which means a shop can find them with a light and a gauge on the day, and both are routinely waved through because the amount of metal involved looks trivial. It is trivial. That is not the point. A toe defect is a notch, and a notch is judged by the sharpness of its root and the stress state of the member behind it, not by how much metal is missing.

Before repairing a toe on an installed member: confirm with the engineer of record or the party responsible for the structure that the member is unloaded or adequately supported before grinding or gouging any of it away, because removing metal from a loaded tension member reduces its section while it is carrying load. Grinding releases the base metal and any coating as respirable dust: stainless and chromate primers give hexavalent chromium under 29 CFR 1910.1026 in general industry or 29 CFR 1926.1126 in construction, old paint can give lead under 29 CFR 1910.1025 or 29 CFR 1926.62, and galvanizing gives zinc oxide. Each is an inhalation route needing capture at the tool or a respirator issued under a written program meeting 29 CFR 1910.134, and none is addressed by eye protection alone. A repair pass is a new weld and carries the arc's ultraviolet, requiring the minimum protective shade from OSHA's filter-lens table at 29 CFR 1910.133(a)(5) and the same fire prevention and fire watch the hot work permit governs under 29 CFR 1910.252(a) or 29 CFR 1926.352 in construction.

The gate, stated once

A weld toe is acceptable when parent metal runs continuously into weld metal through a smooth, obtuse transition. Two things can break that. Undercut removes parent metal at the toe and never replaces it, leaving a groove with the parent surface above it. Overlap deposits weld metal out over parent metal that never melted, leaving a rolled edge with an unbonded crevice underneath it.

Both leave a sharp-rooted notch at the exact line where load transfers between the member and the weld. Neither is judged by lost section. This card owns the toe; the planar-flaw stress concentration inside a joint belongs to the lack-of-fusion card.

What each one looks like

  weld toe profiles, section view

  acceptable undercut overlap
        __                 __                __
       /  \               /  \              /  \
      /    \             /    \            /    \
  ___/      \        ___      \        _______    \
     parent          |  |groove        ^^^^^^^ lip
     runs into       |__|cut below lies on cold
     the weld the parent unmelted parent

Undercut: how it forms

Undercut is melted parent metal that the pool did not flow back into. The arc excavated the toe and the filler arrived somewhere else.

• Current too high for the travel speed. The arc digs a wider, deeper melt than the deposit can refill.
• Arc voltage too high, which widens and flattens the arc cone so its edges reach past the deposit.
• Travel speed too fast, so the pool freezes before surface tension pulls it out to the toe.
• Gun or electrode angle pointed at the vertical member on a fillet, so the melt is biased away from the horizontal side.
• No dwell at the toe on a weave. The weave crosses the toe and comes straight back, giving the pool no time to fill.

It shows up most on the vertical leg of a horizontal fillet, because gravity is pulling the pool the other way.

Overlap: how it forms

Overlap is filler that flowed out over cold parent metal and froze there. It is a wetting failure, and it means the toe never got hot enough to be melted by the arc.

• Wire feed or deposition rate too high for the current, so the pool is large and cool and runs ahead of the arc.
• Travel speed too slow with the arc trailing, so the pool climbs out in front of the arc and lands on unmelted metal.
• Current too low for the joint and the wire size.
• Gun angle too far from perpendicular, pushing the pool sideways onto cold surface.
• A dirty or scaled toe region, which resists wetting even when the energy is right.

Overlap and undercut are produced by opposite errors on the same variable: the balance between where the arc's energy landed and where the filler landed. Overcorrecting one produces the other, which is why a welder who is told to "fill the undercut" and responds by slowing down and adding wire ends up with a rolled toe instead.

Why the acceptance limits are not symmetrical

Undercut has a depth, so a code can allow a small one. Overlap has no depth to measure. It is unbonded metal lying on the surface, so there is no shallow version of it that is any less of a crevice, and code acceptance criteria generally do not permit it at all.

The undercut limits themselves belong to the governing document and they are not one number. AWS D1.1, in the edition your project specification or adopted building code invokes, distinguishes members carrying tensile stress transverse to the weld from those that are not, applies a tighter allowance to the transverse-tension case, and treats thicker material differently. The commonly cited tight limit is 1/32 inch of depth for the transverse-tension condition, and that figure means nothing without naming both the condition and the edition, because the more permissive case in the same clause is larger and is qualified by an accumulated length. Pressure work runs under a different document entirely, and an undercut allowance does not carry between the two.

One gate, two toes on the same connection

A shop welded a beam-to-column connection with fillet welds on both sides of a stiffener. A visual inspection found undercut at two separate toes.

The measurements, taken with a pit-depth gauge across the deepest point of each.

• Toe 1, on the beam flange, where the flange carries tensile stress transverse to the weld: 0.033 inch deep, continuous over about 2 inches.
• Toe 2, on the web stiffener, where the stress is parallel to the weld and the member is in compression under the governing load case: 0.033 inch deep, continuous over about 2 inches.

Identical readings, identical lengths, one weld pass apart.

Correction, printed: a gauge reading is a depth, not a disposition. The 0.033 already contains everything the gauge can tell you, which is depth below the adjacent parent surface. Re-basing it into a disposition requires two facts the gauge does not carry: the stress state of the member behind the toe and its direction relative to the weld, and the edition of the document that sets the limit. Those two are what separate the toes, and no re-measurement changes either.

Toe 1: rejected. Transverse tension puts it in the tight case, and the commonly cited limit there is 1/32 inch, which is 0.03125 inch. The measurement is 0.033. That is over, by about 0.0018 inch or roughly 6 percent of the limit. It is tempting to call 0.033 "about a thirty-second," and that rounding runs in the flattering direction on a member where a notch in tension is the fatigue initiation site, so it is not available. The reading is over the limit and the toe is rejected.

Toe 2: routed, not passed. The parallel-stress case in the same clause carries a more permissive allowance qualified by an accumulated length, and a shop reading a limit for one condition and applying it to the other has skipped the whole point of the clause. The disposition here is to look up the actual allowance for that condition in the adopted edition and check the 0.033 depth and the 2 inch length against both halves of it. This card will not hand over the number, because the number moves between editions and between the structural and pressure documents, and a wrong allowance quoted from memory is worse than a lookup.

The repair, and why it is opposite for the two defects. Toe 1 is repaired by depositing metal into the groove with a small, low-deposition pass and a dwell at the toe, then blending. That is the opposite of the correction for an overlapped toe, which is removal by grinding followed by a pass run with more energy relative to filler. If the same welder had been told only "fix the toes," the instinct that fills an undercut is exactly the instinct that creates an overlap on the next joint.

Check against the sibling rules, with the figures printed. The lack-of-fusion card states that unbonded weld metal is a planar discontinuity generally not permitted by code at any size; overlap in this card is treated the same way, with no allowance offered, which agrees rather than conflicts. The heat-input card states that arc energy compares two procedures only at constant thickness and geometry; this example changes no energy figure and attributes both toes to arc placement and pool behaviour rather than to an energy value. The rounding corollary is applied explicitly and visibly: 0.033 against a 0.03125 limit is over, and it is not written as "about 1/32."

The failure mode. Not on the day. Toe undercut passed as acceptable on a transverse-tension flange is a fatigue crack initiation site, and the crack grows along the toe line under service cycling. Because it starts at the surface it is findable early by magnetic particle inspection on a ferromagnetic member, which is the one piece of good news: unlike a buried fusion flaw, a toe defect stays where an inspector can reach it.

How to verify you got this right

• Measure with a gauge, not by eye. Undercut depth is a small number and a welder's eye at arm's length routinely calls a 0.03 groove "hairline." A pit-depth gauge or a weld gauge with a depth blade takes seconds.
• Write the stress condition on the report next to the depth. A depth with no condition beside it is unusable a week later, and it is the line most inspection reports leave off.
• Name the document and the edition on the disposition, because two editions of the same code can differ on the allowance and a reader will otherwise assume yours is current.
• Check for overlap by running a fingernail or a straightedge across the toe, not by looking at it. A rolled lip and a well-wet toe look similar under shop light and feel completely different.
• Re-inspect after a repair pass, and inspect for the opposite defect. The correction for each one produces the other when overapplied, so a repaired undercut is the highest-probability place in the shop to find a fresh overlap.

References

• AWS D1.1 structural welding code for steel, in the edition invoked by your project specification or adopted building code, for undercut and overlap acceptance criteria and their dependence on stress condition
• ASME Boiler and Pressure Vessel Code Section IX with the applicable piping or vessel code, as adopted by your jurisdiction, where the work is pressure-retaining
• 29 CFR 1910.134 for respiratory protection; 29 CFR 1910.1026 and 29 CFR 1926.1126 for hexavalent chromium; 29 CFR 1910.1025 and 29 CFR 1926.62 for lead in old coatings
• 29 CFR 1910.133(a)(5) for filter shade; 29 CFR 1910.252(a) and 29 CFR 1926.352 for hot work fire prevention during a repair pass
• See related: What Lack of Fusion Is and Why It Hides; What Shielding Gas Is Doing and What Happens When It Stops; What Cracking Tells You About When It Happened