Why a Visual Inspection Finds Most of What Matters

Why this matters

Visual inspection gets treated as the method you use when you cannot afford a real one. It is the opposite. It is the only examination applied to 100 percent of weld length as a matter of routine, while every volumetric method in normal fabrication is applied to a sample, and it is the only one that can run before the joint exists and while it is being filled. Two of its three stages examine conditions that stop existing the moment the groove is full. So the yield of a visual inspection is decided almost entirely by when it happens, and a shop that performs it once, at the end, has kept the name and thrown away most of the method.

Before inspecting anything hot or installed: a joint just welded is at a temperature that burns through a glove on contact, so let it cool or handle it with tongs rather than checking a fillet gauge against it by hand. Do not stand in a welder's line of sight to inspect an in-process joint without the filter shade for that process and current from OSHA's filter-lens table at 29 CFR 1910.133(a)(5), because the ultraviolet burns corneas at distances well past the arc. Inspecting an installed member overhead or at height carries the fall and dropped-object exposures of the access method, not of the weld, so use the fall protection and toe boards the access itself requires under 29 CFR 1926 Subpart M on a construction site. Where a joint has to be cleaned before it can be read, grinding or wire-brushing releases the base metal and its coating as respirable dust, an inhalation route needing capture at the tool or a respirator issued under a written program meeting 29 CFR 1910.134 rather than eye protection alone.

The three stages, and which two disappear

Before welding. Joint geometry, root opening, root face, bevel angle, alignment, cleanliness, and whether the material and consumable in front of you match the drawing and the procedure. Every one of these is measurable in seconds with a gauge and a tape, and every one of them becomes unmeasurable once the second pass is in.

During welding. Root pass condition before it is covered, interpass cleaning, bead placement and sequence, arc strikes on the parent metal, interpass temperature, and any pass that came out wrong while it is still the last thing deposited. Same property: this is a one-time viewing window per pass.

After welding. Profile, size, length, undercut, overlap, surface porosity, cracks, spatter, craters, arc strikes and general workmanship. This is the only stage most shops actually run, and it is the stage where a finding is most expensive to act on.

The first two stages find causes. The third finds effects. That is the whole argument for the timing.

What the method deliberately does not cover

This is the part worth being precise about, because a visual acceptance is often read as a general endorsement of the joint, and it is not. Visual inspection reports on the surface of the weld and its immediate parent metal at the moment it was performed. It says nothing about:

• Anything below the surface. Sidewall lack of fusion, buried slag, subsurface porosity and incomplete penetration on a one-sided joint with no back access are all outside it. The lack-of-fusion card owns why that particular flaw is the one most likely to be sitting there behind a good-looking cap.
• Anything that has not happened yet. Hydrogen-assisted cracking opens hours to days after the arc is out, which the hydrogen card owns. A visual inspection performed while the joint was still warm is a true statement about that moment and an untrue statement about Monday.
• Properties. Hardness, toughness, composition and everything in the heat-affected zone. No amount of looking reaches them, and the heat-affected-zone card explains which document does.
• Material identity. A visual check confirms that a piece is the right size and in the right place. It does not confirm the grade, and two grades of the same family are dimensionally identical.

Being able to say those four sentences out loud is more valuable than any tool on the list below, because it is what turns "we inspected it" into a statement somebody can rely on.

The tools, and the ones that do not belong

A visual inspection kit is small and cheap: a fillet weld gauge set, an undercut and profile gauge, a steel rule and tape, a straightedge, a strong low-angle light, a magnifier of about 5x to 10x, and a temperature indicating crayon or a contact pyrometer for preheat and interpass.

Two items get put in the kit and should not be. A portable hardness tester produces a number with a different basis from a laboratory traverse, on an unprepared surface, and quoting it as a heat-affected-zone hardness misrepresents both figures. A magnifier above about 10x starts resolving surface texture that no acceptance criterion was ever written against, and an inspector who rejects at 30x is rejecting to a standard nobody agreed to. If the question genuinely needs magnification beyond that, the answer is a different method, not a stronger lens.

The low-angle light is the item most often missing and the one that does most of the work: undercut, overlap and a shallow crater are all read by the shadow they throw, and a light held over the inspector's shoulder throws none.

What an acceptance actually records

An acceptance stamp is worth exactly the scope written beside it. "Welds complete, no visible defects" is a statement about defect presence. It is not a statement that the weld is the size the drawing called for, that it runs the length the detail called for, or that it is the joint type specified. Those are conformance questions, and they are answered by comparing the joint against the detail, not by looking for flaws.

Record the stage, the criterion and the scope, in that order: the stage tells a later reader what could have been seen, the criterion tells them what "accept" meant, and the scope tells them how much of the weld was covered. The weld records card owns the field list; the point here is that a visual result without its stage cannot be interpreted afterwards.

Worked example: 12 groove joints on one weldment

An eight-pass groove detail, 12 joints, one weldment, inspected at all three stages against the acceptance criteria named in the contract documents.

Stage one, before welding. Three findings out of 12 joints: one root opening measured 1/4 in against the procedure's 3/16 in maximum, one bevel came in under the specified included angle, and one joint had mill scale left inside the groove. Correcting all three took under half an hour of grinding and re-fitting on the bench.

Stage two, during welding. One finding out of 12: a cluster of start porosity in the root pass of joint 7, ground out and re-run before the second pass went over it. The correction cost one pass.

Stage three, after welding. Two findings out of 12: undercut over the acceptance limit at two toes on joint 3, repaired with a single dressing and re-weld.

Six findings, and the arithmetic that matters is not the count. Take the widest fit-up finding, the 1/4 in root opening on an 8-pass groove. Found at stage one, the correction is a bevel and re-fit, call it 0.2 hours. Found at stage three, the same non-conformance means removing 8 passes by gouging and grinding and depositing 8 passes again, plus re-inspection. The multiple is set by the pass count of the joint, not by the size of the defect, which is why the same finding is trivial in the morning and a day's work in the afternoon. On a two-pass fillet the multiple is small enough that end-only inspection is defensible. On this joint it is not.

What the count does say. Four of the six findings (three at stage one, one at stage two) were found at a stage that no longer existed by the time the cap went on. A shop running only stage three would have found two of six, and the two it found are the two cheapest to fix.

What the stages did not do. Nothing above says the joints are sound below the surface. Joint 7's ground-out root porosity is evidence that the root was watched, not evidence that the sidewalls fused. That question belongs to a volumetric method, and the sibling card on what each method can and cannot see decides which one.

Check the example against the rules the siblings state

• Counts re-derived from the printed list, last mention first. The closing line says four of six findings came from stages one and two: stage one printed three (root opening, bevel angle, mill scale), stage two printed one (start porosity), and 3 plus 1 is 4 against a stated total of 6. Stage three printed two toes on one joint, counted as one finding, and 4 plus 2 is 6.
• The rejected root opening is quoted against its limit. 1/4 in measured against a 3/16 in procedure maximum, which is 1/16 in over. It is rejected in the example, not narrated as marginal.
• The multiple is derived, not asserted. 8 passes removed and 8 redeposited against a 0.2-hour bevel correction, with the pass count named as the driver. No fixed multiple is claimed, and the two-pass fillet case is stated as the condition that reverses the conclusion.
• The delayed-cracking rule is respected rather than assumed away. Nothing in this example is a quenched and tempered steel. Where it were, AWS D1.1, in the edition invoked by the project specification or the adopted building code, requires visual inspection of certain quenched and tempered steels to be delayed at least 48 hours after completion, and stage three would move accordingly.
• No claim of soundness is made from a surface method, which is the same direction the lack-of-fusion and nondestructive-method cards state.

References

• AWS D1.1, structural welding code for steel, in the edition invoked by your project specification or the building code your authority having jurisdiction has adopted, for visual acceptance criteria, inspector qualification, and the delayed inspection requirement for certain quenched and tempered steels
• AWS B1.11, guide for the visual examination of welds, as adopted by your employer's written quality program or invoked by contract
• 29 CFR 1910.133(a)(5) for filter shades, 29 CFR 1910.134 for respiratory protection during weld cleaning, and 29 CFR 1926 Subpart M for fall protection when the access is the hazard
• See related: What Each Nondestructive Method Can and Cannot See; What Lack of Fusion Is and Why It Hides; How Undercut and Overlap Each Form; What a Weld Record Has to Show Later