Third-party review of encoded phased array data

Why phased array is the only common method whose evidence survives

In magnetic particle, penetrant and visual examination the evidence is destroyed by the examination itself. The indication existed on the part for as long as the technician was looking at it, and once the part was cleaned there is nothing left but a written statement. Conventional manual ultrasonics is barely better: the A-scan lived on a screen for a moment, and what survives is a note of amplitude and position. Validation of those methods is necessarily indirect, judging the conditions under which the examination happened rather than the examination itself.

Encoded phased array breaks that pattern. The instrument digitises the full waveform for every focal law and stores it against an encoder position, so the file is not a picture of the examination, it is the examination. A reviewer who never set foot on the site can open it, move the cursor to any point along the weld, and look at the same A-scan the technician saw — plus every A-scan the technician did not stop on, which is usually the more interesting set.

That is why phased array review is a different service from every other line of our report validation work. For other methods the honest deliverable is a judgement about adequacy. For encoded PAUT the deliverable can be a finding about a specific indication at a specific position, defensible because anyone else can open the same file and check it. The distinction matters most when the report is going to be used as evidence.

Scan plan versus what was actually scanned

A scan plan states which focal laws, angles, skips and index offsets were intended to cover which part of the weld volume, and it should be built for the actual thickness, bevel geometry and available scanning surface. The first check is whether one exists as a specific document rather than a template with the job number changed. Generic plans fail on the details that matter: a compound bevel, a counterbore, a thickness transition, or a scanning surface restricted by a support nobody told the planner about.

The second check compares the plan to the acquisition setup stored inside the file. Angle ranges, element apertures, wedge, frequency, index offset and material velocity are all recorded, and they frequently disagree with the plan. A technician who moved the probe out to clear an obstruction and did not update anything has produced coverage that differs from the coverage the report claims, and the file says so even when the report does not.

ASME Section V, Article 4 separates these techniques deliberately. Mandatory Appendix IV addresses phased array manual raster examination using linear arrays, and Mandatory Appendix V addresses E-scan and S-scan encoded linear scanning, where each linear scan runs parallel to the weld axis at a constant standoff with the beam oriented perpendicular to the axis, and the required volume is covered using an encoder. A report that cites one appendix and describes the other is a finding in itself; our notes on ASME BPVC Section V set out how those routes relate.

The encoder record is data, not metadata

Position is the difference between an image and evidence. Without an encoder, a PAUT image tells you what the beam saw but not where the probe was, so nobody can prove the weld was scanned end to end or that a reported indication sits 640 mm from the datum rather than 460 mm. With an encoder, position is recorded at every acquisition step and the whole scan becomes a map that can be checked against the physical part with a tape measure.

Reviewers therefore check the encoder before checking the indications. Was its output verified against a known travel distance before the scan, and again afterwards? Does the recorded scan length match the physical weld length, allowing for run-on and run-off? Does the file show one continuous pass or several restarts, and if it restarts, is there a gap between the end of one and the start of the next? Slippage on a magnetic wheel running over scale is common, real, and visible in the data as a scan shorter than the weld.

Encoder failures also explain a category of dispute that looks like incompetence and is not. When a scan comes up 8% short because a wheel slipped, every indication after the slip is reported at a position that does not match where the client finds it on the part. The client concludes the report was fabricated. The reviewer can demonstrate it is a metrology fault with a defined magnitude and a defined correction, which is a very different finding and a far cheaper remedy.

Proving coverage instead of asserting it

Coverage is the claim that the required examination volume — weld metal, root, cap, and the specified band of adjacent base material — was intersected by beams capable of detecting the flaws the code cares about. Reports assert it in a sentence. Data can demonstrate it. The reviewer overlays the beam geometry from the recorded focal laws onto the weld cross-section and looks for the volume that no beam reached at a usable angle of incidence.

Root coverage is where this usually breaks. A single-sided scan on a thick section, an unfavourable bevel, or an index offset chosen for probe clearance rather than geometry leaves a band near the root reachable only at grazing incidence, where a planar flaw returns very little. That band is precisely where lack of penetration and root cracking live. A coverage overlay showing the gap is far more persuasive to a client than an opinion about whether the scan was adequate.

This is also where the scope of the examining Level III matters. Deciding that a coverage limitation is acceptable is an engineering decision with a defined boundary, discussed in our note on Level III method scope limitations. A reviewer records whether that decision was made, by whom, and whether it was written down before the scan or reconstructed afterwards. The order matters more than the conclusion.

Calibration, TCG and the sensitivity the data was captured at

Every amplitude in a PAUT file is meaningless without the reference it was recorded against. Time-corrected gain compensates for attenuation and beam spread so an equal reflector returns an equal amplitude at any depth, and it is built on a reference block at a specific thickness and material. Used outside the range it was built for, TCG under-compensates at depth, and indications near the far surface are recorded quieter than they truly are — with nothing on the display to announce it.

The reviewer checks what TCG was built on, over what sound path, and whether the examined component sits inside that range. Then the calibration checks. ASME Section V, Article 4 treats a change of 20% or 2 dB in a calibration point as significant: where the response dropped, data since the last valid check is voided and the area re-examined, and where it rose, recorded indications are re-evaluated against the corrected setting. What the reviewer looks for is evidence those checks happened, and what was done on the occasion one failed.

This check is more often exculpatory than damning. A contractor accused of missing a flaw can frequently show, from the calibration record and the recorded amplitudes, that the acquisition was compliant and the indication is genuinely not in the data. That is a finding worth having, and it survives only if somebody kept the file. It is also exactly the question our ASNT Level III consulting engagements are asked to answer long after the crew has demobilised.

Detected is not sized, and acceptance depends on which happened

Detection means an indication exceeded a threshold and was recorded. Sizing means somebody determined its length and through-wall height using a technique whose accuracy has been demonstrated. These are different activities carrying different qualification requirements, but reports blur them constantly: a table with a height column implies sizing occurred, whether or not it did. The reviewer's job is to establish which one the numbers actually came from, using the file rather than the table.

Amplitude drop-out length is not through-wall height. Tip diffraction, mode conversion techniques and dedicated sizing scans produce height. A 6 dB drop along a fixed gate produces a length estimate and says nothing about extent into the wall. Where acceptance criteria are fracture-mechanics based rather than workmanship based, height is the governing dimension, so a height derived from the wrong technique decides acceptance on a number the data cannot support.

ASME Section V, Article 4 addresses this with a dedicated qualification route: Mandatory Appendix IX sets out procedure qualification requirements where flaw sizing — length and through-wall height — and categorisation as surface or subsurface are specified for fracture-mechanics-based acceptance criteria. The reviewer checks which acceptance basis the report claimed and whether a matching sizing qualification exists behind it. Claiming the first without the second is the most consequential defect in PAUT reporting.

Merge, gates and the settings that make indications disappear

Acquisition settings decide what enters the file at all. A recording threshold set high to suppress noise on coarse-grained material also suppresses low-amplitude planar reflectors, and once the scan is over, nothing below that threshold exists anywhere. Unlike gain, which can be re-applied in review, a threshold applied at acquisition is irreversible. The reviewer checks what it was set to and whether the report discloses it, because a report that does not is concealing the limit of its own sensitivity.

Merged views are the other quiet risk. Combining several groups or angle ranges into one displayed image is convenient and entirely standard, but the merge rule — typically maximum amplitude at each point — can present a clean picture while one contributing channel held something worth a second look. Reviewing only the merged view reproduces the original interpreter's blind spot exactly. Opening the contributing channels separately is a routine part of a competent data review, not an exotic step.

Gate configuration completes the set. Gates define which portion of the sound path is monitored and recorded, and a gate that closes before the far-surface return produces a file with no data at all in the region a client later asks about. None of this is exotic or rare. It is ordinary setup, invisible in a PDF, and fully visible in the file. That asymmetry is the whole argument for retaining raw data.

What the codes require, and what they do not

Two questions get confused. The first is whether phased array was permitted for the application at all; the second is what it had to demonstrate once permitted. Construction codes answer the first, and the answer is not universal. Substituting ultrasonics for radiography on pressure boundary welds runs through a specific route, most commonly ASME Code Case 2235, currently at supplement 2235-9, covering Section I, Section VIII Divisions 1 and 2, and Section XII.

That code case matters to a reviewer for a reason beyond permission. Its flaw acceptance criteria were developed using linear elastic fracture mechanics, and its minimum usable thickness was reduced over successive revisions to 1/2 in. (12.7 mm). An acceptance decision made under it is a flaw-size decision, which loops straight back to whether sizing was qualified. A report that cites the code case and then reports amplitude-based acceptance has named a route it did not actually follow.

The second question — what the examination had to demonstrate — sits in ASME Section V, Article 4 and its appendices, plus whatever the referencing code adds. A reviewer states which edition applied on the date of examination, because appendices have moved between editions and criticising a 2017 examination against a 2023 requirement is itself a defective review. Where a client is weighing whether ultrasonics or radiography was the right method to begin with, our comparison of RT and UT for weld inspection covers the trade-offs.

Unprocessed data retention and what a usable raw file contains

A usable archive is the raw acquisition file in the instrument's native format, with setup, focal laws, calibration and encoder data intact — not a screenshot, not an exported image, not a PDF. Exported images bake in the palette, gain and gates in force at the moment of export. They cannot be re-gained, re-gated or re-measured, and every subsequent question about them has to be answered by opinion rather than by evidence.

Retention is a contract term, not a code default, and buyers rarely set it. The party who will need the data years later is the owner, the insurer, or whoever is carrying the risk — and that party is usually not the one deciding what gets archived at project close. Writing a retention period, a format and a handover point into the examination contract costs nothing at the time and is the single cheapest thing an owner can do to keep a future dispute answerable.

The asymmetry is stark. A PAUT file for one weld is measured in megabytes. The cost of reopening a decade-old weld dispute without one is measured in shutdowns. When the file exists, a review costs a day of a Level III's time. When it does not, the same question requires re-examination of a component that may now be insulated, painted, buried, clad or in service at temperature.

When reopening the data confirms the original call

A substantial share of PAUT reviews end by confirming the contractor. The indication is where they said it was, at the amplitude they recorded, the coverage overlay closes, and the calibration record holds. The dispute turns out to be about the acceptance criterion or the contract rather than the examination. Reporting that clearly, with the evidence attached, is more valuable to the party who commissioned the review than a manufactured criticism would ever be.

It also protects the reviewer's usefulness. A review practice that finds fault every time is worth nothing in arbitration, because opposing counsel will produce the pattern and the pattern is the argument. If you are holding PAUT data and a decision that depends on it, send us the file and you will get the answer the data supports. Both directions are deliverable, and both are defensible.

Request a consultation