What a thickness record has to preserve when an LNG plant leaves its legacy system

The migration is a records problem, not a data problem

Most LNG migration projects are scoped as a data transfer: count the rows in the old system, count them in the new one, declare success. That test passes on records that are already ruined. A legacy database that stored a computed corrosion rate but discarded the second-oldest reading will migrate every row intact and still leave you unable to answer the only question that matters, which is how you arrived at the remaining life you published last year.

The useful framing is evidentiary. Each thickness value is a claim about a piece of metal at a moment in time, and its worth depends entirely on the metadata that lets someone else test the claim. Who took it. On what instrument. Using which probe and which mode. Against which calibration block. At what surface temperature. Whether the coating was subtracted. Strip any of those and the value survives as a number while ceasing to be evidence.

LNG plants make this acute because they are long-lived and heavily documented from birth. A train commissioned in 1978 has manufacturer's data records, hydrotest reports, original mill certificates and baseline ultrasonic surveys, most of them in PDF and some of them on microfilm. The new system either becomes the place where a reading joins back to its mill certificate and its baseline, or it becomes another island and the migration has bought nothing.

Cryogenic service changes what a reading interval means

On a warm hydrocarbon line, elapsed time between two readings is a reasonable proxy for exposure. On a cryogenic line it is not. Metal at minus one hundred and sixty degrees is not corroding internally in any practical sense, and the external mechanism runs on the condensation and freeze-thaw behaviour of the insulation system rather than on process chemistry. The damage clock runs during commissioning, warm-ups, upsets and shutdowns, and it runs hard when a vapour barrier has been breached.

This has a direct consequence for the record. A history that stores only date and thickness cannot separate the two regimes, so the corrosion rate is a calendar average across a period that was mostly dormant and partly aggressive. What you need alongside each reading is the service condition at the time of measurement and, ideally, the operating state history of the circuit between readings, so an engineer can compute a rate against warm hours rather than clock years.

API RP 583 treats cold service explicitly and is the right basis for a corrosion-under-insulation strategy on an LNG plant, including the awkward fact that removing insulation and a vapour barrier to inspect can admit moisture and create the damage you were looking for. That trade-off is a documented engineering decision, and it belongs on the CML as an exclusion with a justification, not as an absence.

The tank you cannot gauge, and why its empty CMLs are evidence

A full-containment LNG tank has a 9% nickel steel inner shell inside a prestressed concrete outer container, with perlite and resilient blanket between them. There is no meaningful in-service ultrasonic access to the inner shell. The tank may run twenty-five or thirty years before a planned warm-up and internal entry, and some never see one. This is not a gap in your programme; it is the design.

Legacy systems handled this in two bad ways. Some created no CMLs at all for the inner tank, so a migration inherits an asset with no thickness structure and an auditor concludes the tank was never in scope. Others created CMLs and left them blank for three decades, indistinguishable from a missed inspection. The correct pattern is a populated CML carrying an explicit exclusion class, the technical basis, the accountable engineer and the event that will lift the exclusion, which is normally the next scheduled warm-up.

When that warm-up does happen, the system has to accept a burst of internal readings against CMLs that have been dormant since construction and compute a corrosion rate against a baseline that is thirty years old. That is only possible if the construction thickness, the plate identity and the original survey survived every earlier migration. It is the clearest argument there is for never discarding the oldest reading in a set.

Pulsed eddy current averages are not ultrasonic points

Insulated cold-service circuits are frequently screened without removing insulation, using pulsed eddy current or profile radiography. PEC does not measure a point. It reports an average remaining wall over a footprint that scales with insulation standoff, commonly tens of square centimetres, and it is quoted with a tolerance that is a percentage of wall rather than a fraction of a millimetre. It is an excellent screening tool and a poor trending tool.

The migration trap is mechanical. Legacy exports usually deliver one thickness column, so PEC results and manual ultrasonic results land in the same field. The new system then computes a corrosion rate between a 2019 PEC average and a 2023 UT point and produces a step change of one or two millimetres that reflects nothing except a change of technique. On an LNG plant, where insulation removal is restricted and PEC use is therefore heavy, this can contaminate a large fraction of the cold-service circuits at once.

The fix is to make method a mandatory attribute of the reading and to enforce technique compatibility in the rate calculation. Two readings of different techniques may sit in the same history and be plotted together, but they should not silently define a trend. An integrity engineer who wants to bridge them should have to say so and record why.

Where the warm side of an LNG plant actually corrodes

It is easy to write about LNG as if it were all cryogenic. In practice most of the thickness loss on a liquefaction plant is on the warm side: inlet facilities and feed gas lines carrying CO2 and any H2S, the acid gas removal unit with its amine circuit, regeneration overheads, the dehydration and mercury removal sections, refrigerant loops, fractionation columns and the flare and blowdown system. These are conventional mechanisms and they behave conventionally.

Amine service in particular repays granular history. Rich amine lines corrode at velocity-dependent rates that spike at elbows, tees and downstream of level control valves, and the regenerator overhead sees a different mechanism again. A CML-level record with per-point history lets you see the pattern; a circuit-level minimum tells you only that something somewhere got thinner. Migration decisions that collapse points into circuit minima to save rows destroy exactly the resolution that makes the amine unit manageable.

Regas terminals shift the picture again. Open rack vaporizers put aluminium alloy panels in continuous seawater, submerged combustion vaporizers put flue gas condensate against carbon steel, and the seawater intake system is its own corrosion problem. A thickness history designed only for a liquefaction train will not have the material and environment attributes to make sense of a regas facility, which matters if one system is meant to cover both ends of the chain.

Thickness loss spends your brittle-fracture credit before your pressure margin

On carbon steel equipment in LNG service that can see auto-refrigeration during depressurisation, the governing limit is often not pressure at all. ASME Section VIII Division 1 exempts material from impact testing based on curve assignment and thickness, and UCS-66.1 grants a reduction in minimum design metal temperature based on the ratio of required thickness to actual thickness. The thinner the wall becomes relative to what the pressure requires, the smaller that ratio and the more credit you hold.

The direction of travel is the point. As a vessel corrodes, actual thickness falls, the ratio of required to actual rises, and the MDMT credit shrinks. A vessel that was comfortably exempt at commissioning can drift toward its curve as it wastes, and the failure mode this exposes is brittle fracture on a cold depressurisation, not a leak from general thinning. Remaining-life arithmetic based on pressure alone will never show it.

This is why an LNG thickness history has to retain the specific actual thickness used in any past MDMT assessment, tied to the assessment document, rather than only the current minimum. When someone revisits the calculation five years later, they need to know whether the number in the original report was a point reading, a governing minimum or an averaged value, and which technician and instrument produced it. A migration that keeps only the latest minimum makes the old assessment unreproducible.

Reconstructing a corrosion rate the old system reported

The single most useful acceptance test for an LNG migration is also the cheapest. Pick twenty CMLs across cryogenic, amine and refrigerant service. For each, take the corrosion rate and remaining life the legacy system last published, then ask the new system to reproduce both from the migrated readings alone, with the legacy rate column hidden. Compare.

Expect disagreement, and treat disagreement as the finding rather than the failure. Most mismatches trace to one of a small set of causes: a legacy rate computed against nominal rather than the earliest measured thickness, a t-min that was typed as a constant and never revised after a re-rate, a unit conversion applied once too often, a reading that was superseded in the old system without the supersession being exported, or a rate that was manually overridden by an engineer years ago for a reason nobody recorded. Each of those is worth knowing about before go-live, not after.

Document the reconciliation as a deliverable of the migration, CML by CML, with a disposition for every mismatch. That document is what you show an auditor, an insurer or a new integrity manager when they ask why the numbers changed on the day the system changed. Without it, the migration itself becomes the unexplained step in your history.

How to evaluate the migration before you commit to it

Ask for a pilot on your worst data, not your best. Vendors will happily migrate a clean recent circuit. The circuits that decide the project are the ones from the plant's first decade: inconsistent CML naming, readings in imperial on a metric design, inspector initials only, and rate columns whose inputs are gone. Give the vendor a thousand of those rows and see what comes back, including what the system refuses to import and how it tells you.

Then test the shape of the record rather than the volume. Can it hold two readings on the same CML on the same day from different techniques? Can it store a reading flagged suspect and pending re-shoot without deleting it or letting it drive remaining life? Can it show the full audit trail of a value that was later corrected, including who corrected it and why? Can it reproduce, on demand, exactly what the remaining life was on a given past date, using only the data known at that date?

Finally, test the exclusions. Create a CML that cannot be measured, give it a reason, an owner and a review trigger, and confirm it appears in the inspection plan as a managed item rather than disappearing from every report. On an LNG plant, where a large fraction of the asset is deliberately not gauged, a system that cannot represent a justified absence is not a thickness history at all.

Request a consultation