Welding Defects Guide: All Types, Causes & ISO 5817 Acceptance Criteria

Every fusion weld contains imperfections. The engineering questions are which type, how big, and against what criterion — and that’s what separates a usable defect reference from a photo gallery.

This guide maps all the major welding defect families to one framework: what each defect is (with its ISO 6520-1 code), what causes it, how it’s detected, and how ISO 5817 sentences it at quality levels B, C and D. Each defect links to a dedicated deep-dive with prevention checklists.

Imperfection vs defect — in standards language, every deviation is an imperfection; it’s a defect only when it exceeds the acceptance limit for your specified quality level. The same undercut can be acceptable at Level C and rejectable at Level B. Sentencing happens against a criterion, not against perfection. (How levels are assigned: ISO 5817 acceptance criteria guide.)

The defect map: ISO 6520-1 groups

The defects, one by one

Cracks — never acceptable

The only imperfection family with zero tolerance at every quality level. Three mechanisms with nearly opposite prevention levers: hot cracking (solidification chemistry + bead shape), cold/hydrogen cracking (hydrogen + martensite + stress, appearing up to 48 h after welding), and lamellar tearing (parent plate, through-thickness strain). → Weld cracks: hot vs cold cracking explained

Porosity — the most common volumetric defect

Gas trapped in the solidifying pool. Acceptance scales by level: single pores limited by diameter vs thickness; uniform porosity by projected area (roughly 2% / 4% / 8% at B / C / D). Causes split cleanly into shielding, contamination, and parameters. → Porosity in welding: causes, types, limits

Slag inclusions — the multi-pass tax

Flux residue trapped between passes in MMA/FCAW/SAW. Almost always an interpass-cleaning failure; aligned slag is sentenced severely because it behaves like a planar flaw. → Slag inclusion: causes, prevention, acceptance

Lack of fusion — the inspector’s nightmare

Weld metal touching but not bonded to the joint face or previous pass. Not permitted at any level, hard to detect (RT can miss it; UT/PAUT preferred), and caused by cold parameters or arc placement — the defect that “looks fine” hides best. → Lack of fusion: causes, detection, criteria

Incomplete penetration — the root that wasn’t

The weld doesn’t extend through the designed thickness. Not permitted at Level B for full-pen joints; planar root gap plus undersized throat. Distinct from lack of fusion: penetration is an energy/geometry problem. → Incomplete penetration: causes and acceptance

Undercut — the fatigue starter

A groove melted into the parent metal at the weld toe. The textbook example of level-scaling: h ≤ 0.05t (B), 0.1t (C), 0.2t (D) for the continuous case. Classic symptom of running hot and fast. → Undercut: causes, ISO 5817 limits, prevention

Burn-through — the binary failure

The pool collapses through the joint on thin sections or root passes. Not permitted at any level; on automated lines it’s usually a fit-up variation, not a parameter error — and it has a visible thermal precursor. → Burn-through: causes, prevention, real-time detection · Watch it happen at high speed

Arc strikes — small mark, real damage

A stray arc outside the joint creates a self-quenched, possibly micro-cracked hard spot. Not permitted at B or C; treat as repairable damage everywhere on hardenable steel. → Arc strikes: why they matter

Spatter — the process diagnostic

Ejected droplets; acceptance depends on application (coating, fatigue, hygiene). More valuable as a leading indicator: rising spatter rate means transfer instability — the same instability that produces fusion defects. → Weld spatter: causes and reduction

Shape and dimension imperfections

Misalignment (linear 507, angular 508), excess weld metal (502), overlap (506), and geometric deviations covered by ISO 13920 tolerances. Limits per level are tabulated side by side in the B vs C vs D comparison.

Detection: match the method to the defect

Method selection rules live in ISO 17635; the full comparison including cost-per-detection logic is here: welding inspection methods compared.

The layer post-weld NDT can’t provide

Everything above inspects welds after they’re made — by then, a drifting process has already produced defective metres. The complementary layer is in-process monitoring: thermal imaging and high-speed vision watching every weld as it happens, detecting the conditions that generate defects — shielding loss, cold fusion faces, pool collapse precursors, transfer instability — in real time:

• Real-time weld quality monitoring 101 — how the approach works
• AI defect detection: thermal vs vision vs acoustic — sensor trade-offs
• HeatCore AI — Therness’ production implementation, scoring defect precursors in under 100 ms

In-process monitoring doesn’t replace code-mandated NDT — it reduces the defect population reaching the sentencing gate, concentrates NDT where it matters, and generates per-weld evidence for ISO 3834 traceability.

Quick reference: never-permitted list

For sentencing discussions, the ISO 5817 zero-tolerance imperfections at every quality level:

• Cracks, all types (100) — micro-crack acceptance is material-dependent
• Lack of fusion (401)
• Incomplete root fusion (4013)
• Burn-through (510)

Everything else is a dimensional conversation against Table 1 of the standard — keep a controlled copy at the inspection station, and the level comparison table next to it.