Three LPG Vessels Passed Every Test, Then Failed at 10 Bar

Three LPG storage vessels failed during routine in-service hydrostatic testing in 2024, all below the 23.35 bar test pressure — one at roughly 10 bar — and a UK Health and Safety Executive report, RR1215, traced every failure to brittle fracture that ran around more than 90 percent of the circumference. At manufacture, the production welds had passed 100 percent visual inspection, 10 percent radiography and a hydro-test at 1.25 times design pressure. The inspection was done. The validation never was.

The joggle welds had never penetrated to the backing material. A tandem submerged arc process had laid down a large weld with no grain refinement and minimal cracking resistance. The procedure qualification record on file had been written years after the welds were made, and investigators doubted the procedure had ever been reviewed.

ISO 9001 has required validation of such processes since 2000. Clause 8.5.1 f) of the 2015 edition requires "the validation, and periodic revalidation, of the ability to achieve planned results of the processes for production and service provision, where the resulting output cannot be verified by subsequent monitoring or measurement." The 2026 edition carries the wording across, and its expanded Annex A adds that where the output itself cannot be measured, process validation gives confidence in conformity.

One detail matters: revalidation is tied to changing conditions, not calendar dates. ISO 9000:2015 supplies the term "special process" for outputs that "cannot be readily or economically validated," and ISO 3834-1:2021 applies it to welding by name. Its instruction reads: "Quality cannot be inspected into a product: quality needs to be built in. Even the most extensive and sophisticated nondestructive testing does not improve the quality of the product."

What happened to the requirement's visibility?

In ISO 9001:2000 and 2008, the validation requirement had its own clause and heading, 7.5.2. The 2015 revision folded it into one lettered item in a list of eight. The requirement itself never moved; only the signpost did.

The author of the analysis, a welding professional with 40 years qualifying procedures and welders, writes that no radiograph has ever told them the hardness of a heat affected zone. NDT finds discontinuities large enough for the method to detect, and HSE's own guidance says it "is rarely 100% effective" even at that. It says nothing about strength, toughness, HAZ hardness or the microstructure a given cooling rate leaves behind. Those are proven only on a coupon cut up before production, or during a failure analysis.

Why no NDT method will give you hardness

Weld a hardenable steel and the parent metal beside the fusion line heats into the austenite range and cools at whatever rate the joint dictates. Fast cooling produces martensite: hard, brittle and receptive to hydrogen. Add hydrogen and restraint, and hydrogen-assisted cold cracking follows — often delayed by hours or days, so a weld passes inspection on its shift and cracks after the inspector leaves. Add low toughness and the result is brittle fracture of the kind that took out the LPG vessels.

HAZ hardness is set by heat input, preheat, interpass temperature, joint thickness and steel chemistry — all process variables. Radiography sees density. Ultrasonics sees acoustic boundaries. Neither sees microstructure. The only place HAZ hardness is measured is a polished, etched cross-section on the procedure qualification coupon, against a limit such as ISO 15614-1's 380 HV10 as-welded cap for carbon manganese steels.

RR1215 records a container stacker boom that detached with HAZ hardness readings of 532 and 648; investigators concluded preheat was "either woefully inadequate or non-existent." An ultrasonic test of that joint would very likely have been accepted. The takeaway: you cannot inspect hardness out of a production weld, only control the process within the window the coupon proved.

What validation actually requires

• A qualified procedure. The welding procedure specification, written to ISO 15609-1, ASME IX or AWS D1.1, covers process, consumable, joint, position, parameters, preheat, interpass, heat input and post-weld heat treatment. It becomes validation only when a coupon welded to it passes destructive testing: tensiles, bends, macro, hardness survey and Charpy impacts where the design needs toughness (ISO 15614-1 clause 7, ASME IX QW-451). A file of procedures with no test evidence has validated nothing.
• Revalidation tied to essential variables. ASME IX defines an essential variable as a change affecting mechanical properties; a supplementary essential variable affects notch toughness. ISO 15614-1 sets numbers: heat input held within 25 percent, requalification if preheat drops or interpass rises more than 50 °C.
• Evidence the process ran where qualified. Preheat and interpass measured with a contact thermometer and recorded; consumables controlled by classification and lot; storage per manufacturer recommendations. ISO 3834-2:2021 spends clauses 9 through 14 here and puts inspection before and during welding ahead of inspection after. Welder qualification to ISO 9606-1 or AWS D1.1 proves the welder can deposit sound metal — not that the procedure was right.

Who owns the process?

Someone must decide which procedure applies to a joint, whether a material substitution is acceptable, whether measured preheat matches required preheat. ISO 14731:2019 calls this welding coordination and states in clause 4.2 that "welding coordination is the sole responsibility of the manufacturer." In Australia, AS/NZS 1554.1 clause 4.12 requires a welding supervisor on structural work, with AS 2214.1:2024 setting the qualification. ISO 9001 asks for it in 8.5.1 e).

The author names this role the "key welding person": the person who manages the welding rather than inspects it. In US shops a CWI often fills the gap. ISO 14731 dropped prescribed qualifications in 2019 and replaced them with three competence levels — comprehensive, specific and basic — set against product complexity and application criticality, assessed on experience and theoretical knowledge. In workshops that weld well, this person is findable in five minutes. In shops that struggle, nobody can say who it is, or they live in the office.

What it costs to skip validation

A TWI industry survey puts weld repair rates in oil, gas and power at 1 to 3 percent on average, peaking at 25 percent at specific locations. Each repair means excavation, re-preparation, re-preheat, re-weld, re-NDT and often re-PWHT, on a joint more restrained than the first time.

RR1215 reviewed 47 incidents where weld quality contributed to failure of a safety-critical component: structural collapse in 49 percent, fatalities in 11 percent. Quality issues led at 74 percent, training and knowledge gaps at 38 percent, supervision and acceptance at 28 percent. Welder hand skill sits far down that list; systems and people sit at the top. Aerospace built Nadcap on clause 8.5.1 f) with welding a named special process, and automotive built AIAG CQI-15 — if the clause alone sufficed, neither would exist.

The practical audit: pick one weld shipped last month and trace it — material against drawing, WPS against qualification record, essential variables against production, preheat records, welder currency, consumable lot, and who managed it. Answer these and the process is validated. An inspection report alone shows somebody looked at the final weld, not that it is serviceable. You can inspect a weld; you cannot inspect the welding.