What an independent reviewer checks in a UT thickness report

The readings are not the finding — the pattern is

A thickness survey produces a column of numbers, and the engineering question is never what those numbers are. It is whether they describe the corrosion. A vessel losing wall under a deposit at one seam can return two hundred readings comfortably within tolerance and one reading that was never taken. So the first pass of a validation ignores the values entirely and asks a spatial question: given the fluid, the temperature, the geometry and the operating history, where should this asset be thinnest, and did anyone put a probe there?

That is why the review starts from the damage mechanism rather than the data. Wet H2S service, sulphidation, CO2 corrosion at flow disturbances, erosion downstream of orifices and control valves, dead legs, injection points, soil-to-air interfaces and pipe-support touch points all thin in predictable places. A survey that ignores them is not a survey of the asset, it is a survey of the scaffolding. Across the reports that come through Atlantis report validation, this is the most common single reason a technically clean document fails.

Fairness matters here. Sparse coverage is frequently the client's own doing: a scope that specified a fixed count of readings and said nothing about mechanism, or an outage window that closed before the insulation came off. A reviewer who names the procurement decision instead of the technician produces a finding that survives contact with the contractor's own Level III, and that is the finding the buyer can actually act on.

Grid, spot, and what each one can actually prove

Spot readings answer one question well: how fast is a broadly uniform wall thinning at a point being trended over years. They answer nothing about pattern. Grid scanning and corrosion mapping answer the pattern question — whether the loss is general, patchy, a channel along the six o'clock line, or a cluster of pits — but they generate volumes of data that most programmes never trend, and they consume outage time. Neither is the better technique. They answer different questions, and confusing them is what produces an unusable survey.

So the reviewer's test is a match test, not a preference. If the operator declared pitting, under-deposit attack, erosion-corrosion or microbiologically influenced corrosion, and the contractor delivered single spot readings at fixed CMLs, the survey structurally cannot find the mechanism it was commissioned to find. Conversely, a full corrosion map on a line with uniform, slow, well-characterised general thinning is expensive data with no new information in it, and calling that a deficiency would be wrong.

There is a middle failure worth naming. Some reports use grid language — scan performed, area covered — while the reading log shows four values per location. The narrative and the data disagree. A reviewer resolves that by reading the log rather than the summary, and the finding is usually about report writing rather than fieldwork, which is exactly the category examined in what clients reject in NDT report formats.

CML identity: a number that has to resolve to a place

A thickness reading has no value on its own. Value appears only when the same physical point is measured again, which means the identifier has to resolve to a place. The reviewer takes a sample of CML numbers from the report and tries to find them on an isometric, a shell development or a vessel sketch. If a number cannot be located, the reading is a snapshot rather than a trend point, and every corrosion rate derived from it is arithmetic performed on two unrelated measurements.

Renumbering is the quiet killer. A contract changes hands, the new provider builds a fresh CML list, and CML 12 in the new report is not CML 12 in the old one. Nothing in either document is false. The trend between them is nonsense, and it usually trends in the reassuring direction, because a fresh point on sound metal reads thicker than a legacy point chosen years ago on a corroded one. Validators look for renumbering events by comparing point counts, ordering and reading distributions across surveys.

This is the strongest practical argument for holding thickness history against a spatial model rather than a spreadsheet. When each CML carries a coordinate on the asset, as it does in a digital twin built for thickness trending, a renumbering event becomes visible instead of silent: the new point lands somewhere the old one was not, and the model says so. A spreadsheet cannot detect it, because a spreadsheet does not know where anything is.

Nominal thickness is the input nobody measures

Remaining life is measured loss divided by rate, and loss is measured from a baseline. In most programmes that baseline is a nominal thickness copied from a line list, a drawing or an equipment record — a number nobody measured. The reviewer asks a blunt question: what is the evidence for the nominal, and was it verified or assumed? Reports rarely answer, because the field almost never contains an answer, and the assumption is invisible once it has been carried forward through three inspection cycles.

The tolerance is why this matters. ASTM A106 requires that the wall at any point be not more than 12.5% under the specified wall, and the same minus-12.5% tolerance is standard for ASTM A53 and API 5L. A run of pipe rolled legitimately to the low side starts life below its own drawing nominal. Trend that against nominal and the report attributes a manufacturing tolerance to corrosion, inflating every rate and shortening every remaining life on the circuit.

The error runs both ways, which is why the finding has to be written carefully. Plate and forgings are routinely supplied above nominal, so a vessel course thicker than drawing makes the first survey look like the asset gained metal, and inspectors then discard the reading as an error rather than treating it as a baseline correction. Either way the fix is the same and it is cheap: establish a measured baseline in an uncorroded area and state it in the report, so future reviewers know what the loss was measured from.

Velocity calibration on the actual material, not the block on the bench

A thickness gauge does not measure thickness. It measures transit time and multiplies by an assumed velocity, so a velocity error scales linearly into every reading in the survey. Carbon steel runs around 5,900 m/s longitudinal. Austenitic stainless, duplex, cladding, copper alloys and cast material do not, and coarse or anisotropic microstructures do not have a single constant velocity at all. Setting the instrument on a carbon steel block and then measuring a stainless line biases the entire data set in one direction.

The contact pulse-echo practice adopted as ASTM E797/E797M, and in the ASME family as SE-797 within ASME BPVC Section V, is explicit that it applies where the ultrasonic velocity is constant throughout the part and a back-wall reflection can be obtained and resolved, and it is written for contact measurement at temperatures not exceeding 93°C (200°F). Those two conditions are exactly what gets violated on hot lines and on clad or dissimilar-metal components, and reports almost never mention either.

So the reviewer looks for three artefacts: a setup record naming the velocity used, evidence it was verified on a step wedge or a known thickness of the actual material, and a temperature note where the surface was hot. Where the survey crossed material types, the reviewer checks whether velocity was changed at the boundary. A single velocity applied across a carbon-steel-to-stainless transition is a finding on its face, and it is one of the easiest to prove long after the fact.

Coating: the error that always flatters the asset

Non-metallic coatings transmit sound at roughly half the velocity of steel, so a conventional single-echo gauge times the coating at steel velocity and adds roughly twice the coating's true thickness to the reading. The error is always in the same direction: the steel looks thicker than it is. Ten thousandths of paint can present as more than twenty thousandths of metal that does not exist, and it does so consistently at every coated location in the survey rather than randomly.

Two techniques remove it. Echo-to-echo measures between successive back-wall echoes inside the metal, ignoring the coating layer entirely. Through-coat measurement identifies the round trip inside the coating and subtracts it, returning both the coating and the substrate separately. Both are ordinary instrument functions on current gauges. The problem is not capability, it is that reports routinely fail to state which mode produced which number, so coated and uncoated readings sit in one column and get trended against each other.

That makes coating handling a procedure defect before it is a technician defect. If the written procedure is silent, each technician decides, and the survey silently mixes methods. A reviewer checks the procedure first and the reading log second, and where neither records the mode, the honest conclusion is that coated readings cannot be compared with uncoated ones. Procedure gaps of exactly this kind are what a structured NDT programme audit is designed to surface before the reports are ever written.

Minimum reading, averaged thickness, and which one governs

Minimum thickness means at least three different things inside a thickness report: the lowest reading recorded, the design minimum required for pressure, and the structural minimum. Reports collapse them into one phrase. A validator separates them, because the acceptance rule attaches to a specific one, and applying the wrong definition changes the answer without changing a single measurement. It is the most common source of a dispute in which both parties are reading identical numbers and reaching opposite conclusions.

The codes are specific about averaging, and the specificity is what gets lost. For tank shells, API 653 evaluates a corroded area by taking the least remaining thickness excluding pits, computing a critical length of 3.7 times the square root of the tank diameter multiplied by that thickness, capped at 40 in., and averaging at least five equally spaced measurements over that length. Continued service requires the averaged thickness to meet the minimum, and the least thickness to be at least 60% of it, with the minimum for any shell course not below 0.10 in.

A report that presents only an average hides the point that fails the 60% test. A report that presents only the single lowest reading condemns a shell the code would accept. Both are common, and both are correctable on paper without returning to site, provided the individual readings survive. Where only summary statistics were retained, they do not, and the reviewer's finding becomes one about data retention rather than about the tank itself.

Rebuilding the remaining-life calculation from the raw log

The last stage of a validation is arithmetic, and it is the stage most likely to confirm the contractor. Remaining life is actual thickness minus required thickness divided by the corrosion rate, and API 570 works that way for piping. The reviewer recomputes it from the raw log rather than accepting the printed value, which catches transcription errors, unit slips between inches and millimetres, and required thicknesses taken from a design document that a re-rate had already superseded.

Rate selection is where judgement enters. Two rates exist for any CML with history: a long-term rate from the original baseline and a short-term rate from the most recent interval, and the higher of the two normally governs. A report that publishes one unlabelled rate has made a choice without disclosing it. If that choice was the long-term rate on a CML whose recent interval accelerated, the acceleration has been averaged into a comfortable number and the next inspection date is built on it.

The reviewer then checks the inspection interval that came out the other end. An interval set at half the remaining life is a code convention rather than a result, and it fails quietly when the remaining life it was derived from rested on an assumed nominal. This is the point at which the three earlier findings — location, nominal and coating — stop being technical observations and become a date the operator has been planning around.

What it means when the review upholds the report

Validation frequently confirms the original report, and buyers should expect that outcome rather than treat it as a wasted engagement. In a large share of disputed thickness cases the measurements are sound and the disagreement is about the acceptance rule, the rate that was selected, or the design minimum that was applied. Establishing that in writing is worth as much as finding a defect, because it lets the operator keep the asset in service on a documented basis instead of an argument.

That is also the only honest position to hold in front of an insurer, a purchaser or a tribunal. A reviewer who never confirms anybody's work is not a reviewer, and it shows under cross-examination. If you are holding a thickness report you cannot rely on and cannot re-run, you can send it for independent review and get a defensible answer in whichever direction the data runs.

Request a consultation