Trusting thickness data that arrives from three different contractors

The file arrives as a spreadsheet, and the conditions arrive nowhere

The usual delivery from a third-party NDT contractor is a workbook: tag number, CML number, a thickness value, sometimes a date. Everything that determines whether that value can be compared with last year's value stays behind at the contractor's office — in the technician's field sheet, in the instrument's setup file, in the crew supervisor's memory. Six months later the owner's integrity engineer is asked why circuit 14-P-208 lost forty mils in one cycle, and there is nothing in the record that can answer.

This is the specific failure mode that sends petrochemical buyers looking for a thickness history module, and it is not solved by better trend charts. It is solved by widening what a reading is. A reading is not a number; it is a number plus the conditions that produced it. Store the instrument serial and its calibration certificate reference, the transducer part number and frequency, the measurement mode, the couplant, the surface condition and preparation, the temperature, the applied velocity, the technician and their certification level and expiry, and the raw value alongside any corrected value.

Once those fields exist, the reconciliation conversation changes character entirely. Instead of arguing about whether forty mils is credible, you look at the two records side by side and see that the 2024 crew used a dual-element probe in pulse-echo on a painted surface and the 2025 crew used echo-to-echo after grit blasting. The forty mils is the paint. Nobody has to mobilise a verification crew to find that out.

Hot service breaks the arithmetic before the spreadsheet ever sees it

Petrochemical operations read more equipment hot than almost any other sector. Olefins plants keep cracked gas piping, quench oil circuits, transfer line exchangers and hot amine systems in service continuously, and a five-year turnaround cycle means most thickness monitoring has to happen on-stream. On-stream means elevated temperature, high-temperature couplant, delay line probes and a velocity that is no longer the one printed on the gauge.

Longitudinal velocity in carbon steel decreases with temperature by roughly one percent per 100°F. Some gauges apply a correction automatically if the operator enters a temperature; many do not, and many operators do not. The result is that half your history is temperature-corrected and half is not, and there is no field in the spreadsheet that tells you which half is which. When a later crew starts correcting properly, every corrected location shows an abrupt apparent wall loss that has nothing to do with corrosion. Integrity engineers then spend a shutdown chasing a metallurgical explanation for an arithmetic artefact.

The module has to hold temperature as a first-class field, hold the applied velocity or correction factor, and keep the raw uncorrected reading permanently. Keeping the raw value is what makes a fleet-wide fix possible: if the correction factor used across three years turns out to be wrong, you recompute from raw. If the raw value was overwritten at import, the only remedy is to re-inspect.

Two contractors, two modes, one corrosion event that never happened

Measurement mode is the most common source of false trends in externally supplied data, and it is almost never recorded. Pulse-echo measures from the front wall echo to the back wall echo and therefore includes any coating, paint or scale that the sound passed through. Echo-to-echo measures between successive back wall echoes and therefore ignores the coating entirely. On a lined or heavily painted petrochemical line, the difference is routinely ten to twenty mils.

The pattern this creates is diagnostic once you know to look for it. A whole circuit shifts in the same direction by roughly the same amount in a single cycle, with no localised hot spot, and the shift does not continue in the following cycle. Real corrosion is rarely that uniform and rarely that abrupt. Metal loss driven by sulfidation follows temperature and sulfur content along the circuit; naphthenic acid attack concentrates at high-velocity and turbulent locations; erosion concentrates downstream of elbows and control valves. A flat, uniform, one-cycle step is a method change.

There is a second mode trap worth building a validation rule around. Every gauge has a minimum measurable thickness, and below it many will report double the true wall by locking onto the second back wall echo. On instrument air tubing, small-bore sample lines and thin utility piping, a doubled reading looks like a suspiciously healthy component. A history module that flags any reading close to twice a nominal wall, or twice the previous value, catches this before it reaches a remaining-life calculation.

The exclusion record is the finding, not the footnote

On a live petrochemical unit, a meaningful proportion of scheduled CMLs never get read on any given campaign. The scaffold came down early. The insulation was not stripped because the strip and reinstate order was never raised. The surface temperature exceeded the couplant rating. The location sits inside a fireproofed section, behind a pipe support, or under a shoe where the probe cannot make contact. Access was blocked by an adjacent contractor's work front.

Each of those is a different problem with a different owner and a different cost. Recorded as a blank cell, they become indistinguishable and therefore invisible. Recorded as coded exclusion reasons attached to the CML, they aggregate into something useful: this unit has forty-one CMLs excluded for insulation strip, which is an inspection-planning purchase order; nine excluded for surface temperature, which is a procedure and consumables problem; six excluded because the CML was never physically locatable, which means the CML register is wrong and should be corrected rather than carried forward for another decade.

There is a safety argument as well as a planning one. Coverage that appears complete but is not is how an unexpected failure gets to happen on a circuit everyone believed was monitored. The exclusion rate per circuit, trended over campaigns, is one of the few leading indicators of integrity data quality that costs nothing to produce. A circuit whose exclusion rate climbs every cycle is a circuit whose apparent stability is manufactured.

What the PSM file is required to contain anyway

Under OSHA 29 CFR 1910.119(j)(4)(iv), each inspection and test performed on process equipment must be documented with the date, the name of the person who performed it, the serial number or other identifier of the equipment on which it was performed, a description of the inspection or test, and the results. That is a provenance requirement written into federal law, and it maps almost field for field onto what makes contractor data trustworthy. If your thickness history satisfies the regulation properly, most of the data-quality problem is already solved as a side effect.

The paragraph before it, 1910.119(j)(4)(ii), requires that inspections and tests follow recognised and generally accepted good engineering practice — in this context API 510, 570, 574 and 653, and ASME Section V for the examination method itself. Naming the governing code and edition on the inspection record matters because retirement thickness, interval and the treatment of localised versus general loss all differ between codes, and codes are revised. A reading assessed against the 2016 edition of a code and re-assessed against the current one should carry both determinations, not have one silently replace the other.

The interaction that catches people out is management of change. A metallurgy upgrade, a service change, a re-rate or a new corrosion inhibitor programme changes what the historical readings mean, but the readings themselves are unchanged and keep trending as if nothing happened. Linking the MOC record to the affected CMLs, so that a corrosion rate calculation can be told to restart its baseline at the change date, is the difference between a rate that describes the current process and one that averages two different processes together.

Damage mechanisms that only appear in comparable data

The mechanisms that matter most in petrochemical service are precisely the ones that noisy data hides. Sulfidation in hot hydrocarbon streams follows the Couper-Gorman relationship between temperature, sulfur content and silicon content of the steel, and produces smooth general loss that can be mistaken for measurement scatter until several cycles are stacked. Carburisation and metal dusting in ethylene furnace and syngas service produce localised loss with embrittlement that a thickness reading alone will not characterise, so the reading has to point to the supporting examination that did.

High-temperature hydrogen attack is the case where thickness history is genuinely the wrong tool and knowing that is itself the finding. HTHA is fissuring and decarburisation, screened by operating position against the API 941 curves and confirmed by advanced ultrasonics — typically TFM or backscatter techniques — not by wall thinning. A history module that records the method against every reading lets an engineer answer immediately whether a hydrogen-service vessel has ever been examined for HTHA, or has merely been thickness-monitored for twenty years and assumed to be covered.

Polythionic acid stress corrosion cracking is a third case with a timing dimension. It develops in sensitised austenitic stainless steel during shutdown when sulfides, moisture and air combine, which is why AMPP/NACE SP0170 concerns neutralisation and purging practice at shutdown. Whether that practice was followed on a given outage belongs on the equipment history next to the inspection data, because it determines what the next campaign should be looking for and with which method.

How to evaluate the module rather than the brochure

Evaluate on ingestion first, because that is where petrochemical data quality is won or lost. Hand the vendor two genuine contractor deliverables in their native formats, ideally from different contractors, and ask them to import both in front of you. Watch specifically for four behaviours: whether units are inferred rather than declared, whether unmapped columns are dropped silently or held for review, whether the original file is retained and retrievable from any reading it produced, and whether the import can be reversed as a unit when it turns out to be wrong.

Then test the argument path. Pick any reading in the demonstration dataset and ask the system to show you the technician, their certification status on that date, the instrument and its calibration certificate, the mode, the temperature, the raw value and every edit made since. If the answer requires opening a second system or a shared drive, the module does not do the job you are buying it for. Ask the same question of an excluded reading, because that is where most systems have no answer at all.

Finally, test governance. Ask whether the system can produce an exclusion rate and a rejected-reading rate per contractor, per crew and per circuit over the last three campaigns. That single report changes the contractor relationship from disputes about individual numbers to a measurable quality conversation held at contract review. For a working session on how your existing contractor files would import and what provenance survives, request a demonstration at info@atlantisndt.com.

Request a consultation