Findings, NCRs and Weld Repairs That Survive an Audit

In a fabrication shop the deficiency is a weld, not an asset

Plant-side systems attach findings to equipment: a vessel, a line, a tank, a circuit. That model breaks in a shop, because the object under examination does not exist as an asset yet. It is a joint on a drawing, inside a router package, and it will not carry a serial number or a nameplate until it is stamped and shipped. A deficiency register built on an equipment tree has nowhere sensible to put a rejected root pass on Joint 14 of Spool SP-208-04.

The consequence is practical rather than philosophical. A weld carries its own set of parents: the WPS it was made to, the PQR behind that WPS, the welder's stamp, the filler metal heat and lot, the base material heat, the NDE procedure, the examiner, and the acceptance criteria the customer's specification actually invoked. The moment the weld is rejected, every one of those parents becomes evidence. Attach the finding to a future asset number and you have discarded all of it at the first click.

This is why fabrication deficiency tracking behaves more like genealogy than like a maintenance backlog. The primary key is a joint identifier that is unique across the job, stable across drawing revisions and printable on the traveler. The rejection, the disposition, the repair procedure, the re-examination and the final acceptance all hang from that one node. Get the key wrong and no amount of workflow on top will make the record reconcile later.

The audit that is actually coming is your own quality manual

Three different audits arrive at a fabrication shop and they do not ask the same question. The Authorized Inspector works hold points continuously and cares whether the paper in front of them at that moment is complete and current. The client's QA surveillance team pulls a sample of travelers and traces them end to end, looking for a chain that breaks. The triennial ASME survey with the Authorized Inspection Agency and the Society reviews whether the shop does what its Quality Control Manual says it does.

The last one is the audit shops lose, and it is the one they prepare for least. The QC Manual is a document the shop wrote, submitted and signed, and it is the standard the shop is measured against, not a generic notion of good practice. If the manual states that rejected welds are dispositioned on a numbered NCR, cross-referenced to a repair procedure and re-examined to the original extent, that is the chain that must exist on every rejected weld. Doing something reasonable but different is a finding, and it is a systemic one.

A register earns its place when it makes the manual's route the only available route. If the manual requires a repair procedure reference before a disposition can be set, the disposition should not save without one. If it requires re-examination at the original extent, the closure should ask for the extent and compare it. Systems that let a coordinator type 'repaired and accepted' into a free-text field will reliably produce the exact audit finding the manual invited.

Repair rate: the arithmetic that starts arguments

Nearly every client specification sets a weld repair rate threshold, and nearly every one defines it differently. Rejected welds divided by welds examined. Rejected film length divided by total film length. Rejected exposures divided by total exposures. Repairs divided by welds completed, examined or not. The same shop, the same week, the same welds, produces four numbers that are not close to one another, and each of them can be defended.

The trap is the denominator, and it bites hardest on jobs running partial radiography. Under spot radiography per UW-52, the examined population is a small and deliberately non-random fraction of the welded population, so a rate computed on welds examined moves several points on a single bad shot. The same rejection expressed over welds completed disappears into the noise. Neither figure is dishonest. They answer different questions, and the contract names one of them, usually in a sentence nobody re-reads until the surveillance visit.

A register that stores only a reject flag cannot reproduce either figure. It has to carry examined length, exposure count, rejected length and repair count as separate quantities, so the definition the specification names can be computed on demand rather than negotiated in a conference room. When the number is challenged mid-project, the shop that can display the arithmetic and the underlying joints gets its afternoon back; the shop that cannot spends a week rebuilding it from film logs.

Welder continuity and the lapsed-qualification trap

ASME Section IX, QW-322, expires a welder's qualification for a process after six months without using that process, unless continuity is maintained by one of the permitted routes. Every shop knows the rule. Almost no deficiency register knows it, because the register carries dates while qualification state lives in a separate continuity log or in the QC coordinator's memory.

The exposure is specific to repairs, and that is what makes it hard to see. The original weld is made while the welder is comfortably current. The film is read, the rejection is raised, the spool is set aside for engineering disposition, and the repair happens eleven weeks later, or after a shutdown, a reassignment, or a stretch on a different process. The repair is a new weld, and the qualification that matters is the one standing on the repair date. It will pass the re-shoot and it will still be a nonconformity that traces straight out of a ten-traveler sample.

The remedy is unglamorous. The deficiency record has to resolve the assigned welder's qualification state as of the repair date, not the original weld date, and refuse the assignment when the continuity clock has run out. The same interlock applies to the examiner: SNT-TC-1A certification must be current on the date of examination, including the re-examination. Two date-resolved checks remove a whole category of corrective action from every survey cycle.

Drawing revisions and identity that has to hold

Weld numbers are not permanent. Isometrics get revised, spools get resplit when a shipping envelope changes, a field weld becomes a shop weld when the schedule slips, and the numbering renormalizes across a drawing. Findings raised under the old numbering do not renumber themselves, and nobody notices at the time because the physical weld is sitting right there and everyone knows which one it is.

What surfaces months later, usually in front of a client auditor, is a repair radiograph whose film identification matches no weld on the current drawing. The weld exists. The repair was competent. The film is readable and acceptable. The package simply does not reconcile, and explaining that in an audit costs more than the repair ever did, because a documentation break of that kind invites a wider sample.

A register that survives this keeps an internal identifier that never changes and treats the printed weld number as an attribute of a specific drawing revision. Every finding then carries the revision under which it was raised, and issuing a new revision produces an explicit reconciliation list — joints added, removed, renumbered, and the open findings attached to each — instead of silent drift that only becomes visible when the data book is assembled.

Aging a finding in shop days against a ship date

Plant deficiency backlogs age in months against a risk ranking, because the equipment is in service and the question is how long the condition can be tolerated. A shop's backlog does not work that way. A finding is old when it threatens a ship date, and the clock that matters counts working days remaining before the spool has to leave the blast and paint bay, not calendar days since the film was read.

Sequencing sharpens this further. A weld rejection discovered after post-weld heat treatment is a different animal from the identical rejection found before it, because the repair reopens the heat treatment and, depending on the code case and the customer's specification, the hardness survey and the hydrostatic test as well. A register that ranks findings by indication severity alone will cheerfully leave a minor porosity call sitting behind a PWHT gate until it becomes the most expensive line item on the job.

Useful aging in a shop is therefore positional rather than temporal: how many process gates remain downstream of this finding, and which of them does the repair force you to repeat. That is a computable field once the router is modelled, and it changes the daily production meeting more than any severity label ever has. The item that jumps to the top is rarely the worst indication; it is the cheap one about to be buried.

What the data package has to produce on demand

At the end of the job the shop hands over a document package. Where a Code stamp applies, that includes the Manufacturer's Data Report; alongside it sit the customer's document requirements — weld maps, NDE reports, material test reports, PWHT charts, welder qualification lists, procedure qualification records and the nonconformance history.

The nonconformance history is the section assembled by hand at eleven o'clock the night before the package ships. It is assembled by hand because the rejections live in the NDE reports, the dispositions live in a QA folder, and the re-examinations live in a second set of NDE reports numbered independently of the first. Nothing connects the three except somebody's recollection, and that person is usually on another job by then.

The measurable test of a deficiency system in a fabrication shop is whether that section prints itself, correctly, with no manual reconciliation. If it does, the same underlying query answers the Authorized Inspector's hold-point question, the client's surveillance sample and the survey team's manual-compliance check, because all three are asking for the same chain from three directions.

How to evaluate a system before you commit to it

Bring a real closed finding to the demonstration, and choose one that went badly. A repair on a weld that was later renumbered. A rejection found after PWHT. A repair welded by someone whose continuity had lapsed and was requalified mid-stream. Vendors demonstrate on clean data; your data has never been clean, and the difference is the entire value of the exercise.

Then ask four narrow questions and watch for hedging. Can a finding be raised against a joint that does not yet exist as an asset? Does closing a finding require re-examination evidence at a stated extent, and does the system compare that extent to the original? Is welder and examiner certification resolved against the event date rather than today's date? Can the system produce the client's repair rate definition, not just its own?

If the answers are yes, the rest is configuration: mapping your Quality Control Manual's route into the workflow so the software cannot let anyone off it. That mapping is the work, and it is worth doing with someone who has sat on the other side of an ASME survey. To walk through a live example against your own manual and a job you have already shipped, request a working session at info@atlantisndt.com.

Request a consultation