Key takeaways
- On 3 mm plate, ISO 5817 level B permits no continuous undercut at all, level C permits 0.3 mm and level D permits 0.6 mm, so the level must be written on the drawing or the acceptance criterion does not exist.
- On 10 mm plate, level B and level C both cap continuous undercut at 0.5 mm, which means specifying C instead of B buys nothing on that imperfection.
- Level B allows 1.0% porosity by projected area on a single-layer butt weld, so a 100 mm seam with an 8 mm cap may hold about ten 1 mm pores; level C allows fifteen and level D about twenty-five.
- Use level D for static, non-sealing joints at 3 mm and above, level C as the default for coated load-bearing frames, and level B only for fatigue, sealing or visible undressed faces.
- ISO 5817 does not set the extent of inspection; pair every level with a test method and percentage from ISO 17635 and the contract, or the level applies only to what one inspector can see.
A drawing that says only "weld to ISO 5817" gives three inspectors three different answers. On 6 mm plate, level B allows no continuous undercut at all, level C allows 0.6 mm and level D allows 1.0 mm. The standard is a scale of workmanship, not a pass mark, and a scale without a setting is not a specification.
ISO 5817:2023, the fourth edition, sets limits on imperfections in fusion-welded joints in steel, nickel and titanium alloys. It applies from 0.5 mm thickness upward, covers fully penetrated butt welds and all fillet welds, and refers every imperfection back to an ISO 6520-1 number. The operative content is one table: thickness bands on the left, imperfection types down the middle, three columns of limits for levels D, C and B.
A usable weld specification has three parts: the level, the extent and method of inspection, and a note on which faces are visible. The sections below give the limits that decide most arguments, the arithmetic behind them, and where the standard hands the decision back to you.
What ISO 5817 defines, and what the levels mean
The standard defines quality levels for production quality and good workmanship, and its scope says so in plain terms. The levels are not fitness-for-purpose. They describe how well a joint was made, not how much load it carries. A fatigue-critical bracket welded to level B can still crack at the toe in a few hundred thousand cycles, because the joint geometry, not the workmanship, sets the stress concentration.
Three levels exist. B is the strictest, D the loosest, and C sits between them and is where most general fabrication lands. Each imperfection has its own limit in each level, and the limits change with thickness. That second point matters more than the first. The same 0.5 mm undercut passes at level B on 10 mm plate and fails at level B on 3 mm plate, because level B does not permit continuous undercut at all below 3 mm.
Because everything is keyed to ISO 6520-1 numbers, a drawing note can be precise without copying the table. "Undercut (5011) to 0.2 mm maximum, porosity (2011, 2012) to 1.5% of projected area" is a complete instruction that a fabricator, an inspector and a third-party all apply the same way. Two rows copied onto the drawing usually beat one word on the drawing.
The standards and compliance hub collects the scope notes and inspection guides that sit around this subject.
The limits that decide most arguments
Four imperfection families cover most shop-floor disputes on sheet-metal fabrications: cracks, undercut, porosity and excess weld metal. The values below come from Table 1 and apply to butt welds unless marked otherwise, where t is nominal thickness and b is weld width.
| Imperfection (ISO 6520-1) | Level B | Level C | Level D | Ganesh limit or note |
|---|---|---|---|---|
| Crack (100), crater crack (104), t ≥ 0.5 mm | Not permitted | Not permitted | Not permitted | Grinding out a crack is a repair weld with its own WPS, not a cleanup |
| Continuous undercut (5011), t 0.5–3 mm | Not permitted | ≤ 0.1t | ≤ 0.2t | On 1.2–2 mm canopy panels, level B on undercut becomes a grinding operation |
| Continuous undercut (5011), t > 3 mm | ≤ 0.05t, max 0.5 mm | ≤ 0.1t, max 0.5 mm | ≤ 0.2t, max 1.0 mm | Above 5 mm plate the cap makes B and C identical, so C is free |
| Uniformly distributed porosity (2011, 2012), projected area, single layer | ≤ 1.0% | ≤ 1.5% | ≤ 2.5% | Only visible on a radiograph or a macro; a visual check cannot grade it |
| Linear porosity (2014), butt weld | h ≤ 0.2s, l ≤ s, max 25 mm | h ≤ 0.3s, l ≤ s, max 50 mm | h ≤ 0.4s, l ≤ s, max 75 mm | The one that shows up in a water leak test on a sealed enclosure |
| Excess weld metal, butt (502), t ≥ 0.5 mm | h ≤ 1 + 0.1b, max 5 mm | h ≤ 1 + 0.15b, max 7 mm | h ≤ 1 + 0.25b, max 10 mm | On a visible face this is a dressing callout in disguise; dress and grade D |
| Incomplete penetration (402), partial-penetration butt | Not permitted | Not permitted | h ≤ 0.2t, max 2 mm | If the joint must carry through, draw full penetration; no level converts a partial joint |
| Multiple imperfections in one cross-section (4.3) | Sum h × l ≤ 4% | Sum h × l ≤ 8% | Sum h × l ≤ 16% | Two individually acceptable imperfections can still fail as a group |
Two patterns in that table repay the reading time. Cracks are not negotiable at any level; a crack is a defect, not a quality choice. And the caps do most of the work on thicker material, which is why a level C note on 10 mm plate is often indistinguishable from level B on the imperfections people actually measure.
Worked limits on 3 mm and 10 mm plate
Undercut is the clearest place to run the numbers, because everything depends on the thickness band. On 3 mm CRCA, level B permits no continuous undercut, level C permits 0.1 × 3 = 0.3 mm, and level D permits 0.2 × 3 = 0.6 mm. On 10 mm plate, level B permits 0.05 × 10 = 0.5 mm, level C permits 0.1 × 10 = 1.0 mm but is capped at 0.5 mm, and level D permits 0.2 × 10 = 2.0 mm capped at 1.0 mm.
Porosity is graded as an area fraction, not a pore count, which is where inspection reports go wrong. Take a 100 mm run of butt weld with an 8 mm wide cap: the projected area is 800 mm². At level B, 1.0% is 8 mm² of pore area. A 1 mm diameter pore covers about 0.79 mm², so the seam passes with roughly ten pores of that size in 100 mm. Level C gives 12 mm², about fifteen pores. Level D gives 20 mm², about twenty-five. The same physical seam is accepted or scrapped on one character in a drawing note.
That arithmetic shows why level B is a process decision rather than an inspection decision. Holding ten pores per 100 mm on 1.5 mm sheet means controlling gas flow, wire condition, travel speed and fit-up gap. On a robotic cell the parameters repeat and the count is stable. On a manual station the same joint varies between welders and across a shift.
Excess weld metal runs the other way, because its limit depends on weld width rather than load. A 6 mm cap at level B is limited to 1 + 0.1 × 6 = 1.6 mm of reinforcement, at level C to 1.9 mm, and at level D to 2.5 mm. On an undressed face inside a canopy, 2.5 mm is a legitimate level D and an unnecessary reject as level B.
Choosing a level: the decision rule
Use level D when the joint carries static load only, sits at 3 mm or thicker, does not have to seal and is not on a visible face. Use level C as the default for load-bearing frames, enclosures and canopies that will be coated, vibrated or handled, which covers most of the output of a sheet-metal shop. Use level B when any one of four conditions holds: the joint is fatigue-loaded, it must seal against water or air, its face is seen and not dressed, or a downstream part depends on the weld profile, such as a gasket seat or a rail that slides.
Two traps sit beside that rule. Raising the whole drawing to B to be safe adds toe dressing, full visual inspection and volumetric sampling to welds nobody will ever load, and it trains the shop to argue about 0.1 mm instead of about fit-up. Dropping everything to D is worse: a 3 mm D fillet passes inspection and still leaks at the first monsoon, and the drawing gave the fabricator no reason to know the difference.
Where fatigue governs, Annex B of the standard adds criteria and a designation to match. ISO 5817-B90 applies the level B limits with adjustments for a FAT 90 detail, and C63 does the same against FAT 63. Using those suffixes is more honest than writing B and implying a fatigue life the standard never calculates.
Where ISO 5817 stops, and what replaces it
The standard does not say how much inspection to do. That link runs through ISO 17635, which correlates quality level, test method and acceptance level, and then through the purchase order. The standard also does not define detection. Radiographic technique is ISO 17636-1, ultrasonic testing is ISO 17640 with acceptance to ISO 11666, magnetic particle is ISO 17638 with acceptance to ISO 23278, penetrant is ISO 3452-1 with acceptance to ISO 23277, and radiograph acceptance is ISO 10675-1.
It does not apply to aluminium, which has its own scale in ISO 10042, or to beam welding, which is ISO 13919-1. It does not cover fit-up and dimensional deviation, which ISO 13920 handles. And its scope excludes metallurgical properties such as grain size and hardness, so a heat-affected zone that hardened past the parent specification sits outside the level system entirely.
Product standards override it where they exist. A lift component is judged against the lift standard, a railway part against the railway specification, and pressure equipment against its design code. In those cases ISO 5817 is the workmanship baseline, not the acceptance authority, and a supplier arguing only from the standard is arguing from the wrong document.
What replaces the missing parts is a weld map: joints identified on the drawing, a level per joint, a method and extent per joint, and a repair route. For a 2 m frame that is six to ten lines on the drawing, and it removes the argument at first article.
What Ganesh runs on a welded assembly
Repeat welded assemblies run on three CLOOS robotic MIG cells with fixtures designed and built in-house. The robot's contribution to weld quality is repeatability: the same travel speed, the same stick-out and the same sequence on every part, which is what holds porosity and undercut inside level C without a repair loop. More than 30 manual MIG/MAG, spot and special-purpose stations take the larger frames, the low-volume work and the joints a robot cannot reach.
Fixtures are built and checked before a run because fit-up gap is the largest single cause of both undercut and excess penetration. Material is cut on a 12 kW fibre laser with a 40 ft × 16 ft bed or formed on the 200 T, 5-axis brake with its 4,450 mm bed, so parts reach the fixture inside the ±0.5 mm sheet-metal tolerance before welding begins.
Certification is ISO 9001:2015 and ISO 14001:2015. PPAP and first article inspection are available, and a first article for a welded assembly can carry the weld map, the visual inspection record and the dimensional report together. Where the drawing calls for radiography or ultrasonic testing, raise it at RFQ stage: it changes the route, the cost and the inspection hold points, and it is not a detail to discover at dispatch.
Customer work in this shape includes elevator frames for TKE and Eros Elevators, and base frames and canopy structures for Mahindra Powerol. The welded assemblies capability sets out the cell and station range, and welding inspection and traceability covers the record set behind it.
What no quality level fixes
A level grades a weld that the drawing already made possible. It cannot fix a 3 mm fillet called on 0.8 mm sheet, where the arc burns through before the toe forms, and it cannot fix a joint that needs a backing bar no one can reach. Both failures are design problems that a level B callout makes louder rather than better.
Distortion is the other case. Weld shrink moves a frame by 0.5–1.0 mm per fillet, and no undercut limit addresses it. That is a fixture and sequence question, covered in welded assembly tolerances, and it usually matters more to a fitter on site than the porosity grade.
What to put on the drawing
- The level per joint family, written as ISO 5817-B, -C or -D, plus one stated default for joints not listed.
- The FAT suffix where fatigue governs, for example ISO 5817-C63 or ISO 5817-B90.
- The imperfection numbers that matter for the part, with tighter limits written out, such as undercut 5011 to 0.2 mm on a sealing face.
- The method and extent: visual to ISO 17637 on all production, plus radiography or ultrasonic testing on a stated percentage of a stated joint.
- The acceptance document for indications, so the NDT supplier and the fabricator grade against one scale.
- The faces that are dressed, the faces visible after coating, and the spatter requirement if powder or paint follows the weld.
- The repair rule: who approves a repair WPS, and whether a repaired area is re-tested to the same level.
- The thickness range covered, because the limits move with t and a substitution from 3 mm to 2 mm is a different acceptance case.
Send the drawing with the weld map attached, or the joints that keep failing at first article marked up, through the quotation route. The engineering team returns the level, the inspection extent and the process change behind each callout, and the robot versus manual welding comparison explains which cell the work would run on.
Frequently asked questions
Does "weld to ISO 5817" on a drawing mean anything on its own?
Not enough to inspect against. The standard has three quality levels and the limits differ by thickness, so a drawing needs ISO 5817-B, -C or -D plus the test method and extent. Without a level, two inspectors can both be correct and disagree.
Which ISO 5817 level should I specify for a sheet-metal enclosure?
Level C is the usual default for a coated enclosure in 1.5–3 mm steel. It permits 0.1t undercut, capped at 0.5 mm, and 1.5% porosity by projected area, which a robotic MIG cell holds without a repair loop. Move to level B only on sealing joints, visible undressed faces or fatigue details.
Is a level B weld stronger than a level D weld?
Not by definition. The standard's scope states that the levels describe production quality and workmanship, not fitness for purpose. Strength comes from the joint design, the throat thickness and the load path, so a B weld on an undersized fillet is still an undersized fillet.
What does ISO 5817 say about spatter?
Level B does not permit spatter on the finished weld; levels C and D allow sporadic spatter where it does not affect function. For a part that will be powder coated, spatter is a pretreatment and adhesion problem, so write a spatter limit even when the weld is graded D.
Which standard covers weld quality in aluminium?
ISO 10042 covers aluminium and its alloys with the same B, C and D structure. ISO 5817 itself is written for steel, nickel and titanium alloys, so quoting it on an aluminium canopy does not give the fabricator a valid acceptance criterion.
Can a fabricator repair a weld that fails the stated level?
Yes, with an approved repair procedure, because a repair is a new weld with its own heat input and its own risk of cracking. Ask who approves the repair WPS and whether the repaired area is re-tested to the same level; on level B work the answer should be yes to both.
Sources and further reading
- ISO 5817:2023 — Fusion-welded joints in steel, nickel, titanium and their alloys (beam welding excluded): quality levels for imperfections, levels B, C and D for material 0.5 mm and above · reviewed 3 Nov 2026
- ISO 2553:2019 — Welding and allied processes: symbolic representation on drawings, the source of weld size, length and finish on a fabricated part · reviewed 3 Nov 2026
- ISO 3834-2:2021 — Quality requirements for fusion welding of metallic materials, part 2: comprehensive quality requirements, which controls the process that produces the weld · reviewed 3 Nov 2026
Have a weld drawing with no quality level?
Send the drawing and the joints that fail at first article. We return a weld map with a level, a test extent and the process change behind each callout.
Request a weld review