Making Contractor Thickness Data Trustworthy in an Ammonia Plant

Why the readings arrive from someone else

Ammonia and fertilizer complexes are capital-heavy and headcount-light. A world-scale synthesis train, a urea melt and granulation plant and a nitric acid unit are often covered by one integrity engineer and one mechanical inspector, with all UT, RT and phased array bought from regional contractors on annual frame agreements. The agreement is retendered every three to five years, and every retender resets the conventions under which your corrosion rates were built. Nobody notices, because the numbers keep arriving in the same spreadsheet template with the same column headings.

The consequence surfaces at the worst possible moment. API 510 sets the next internal inspection at the lesser of half the remaining life or ten years, and remaining life comes from a corrosion rate derived from two thickness values. When those two values were produced by different firms using different measurement modes on differently prepared surfaces, the rate is partly an artefact of your procurement calendar. A rate forty percent too high pulls a shutdown forward and burns capital you did not need to spend. A rate forty percent too low gives you an interval you cannot defend when someone asks how it was derived.

The module exists so that provenance travels with the number instead of living in the covering report that nobody opens once the campaign closes. Every field that could explain a discrepancy is attached to the reading row itself, which means it survives export, merge, migration and the next change of contractor.

The damage mechanisms that punish a flattened record

Fertilizer complexes carry an unusually wide spread of mechanisms in a small footprint. Refrigerated anhydrous ammonia storage tanks are susceptible to ammonia stress corrosion cracking, which is sensitive to oxygen contamination and water content and is a cracking problem rather than a thinning problem, so thickness history has to sit alongside crack-detection results rather than replace them. Wet CO2 corrosion in the carbon dioxide removal and stripping circuits is aggressively local and gravity-driven. Carbamate corrosion in urea high-pressure equipment is a passivation problem: lose the passivating air and stainless that has behaved for a decade begins losing wall at rates that make an annual inspection interval meaningless.

Some of what threatens these units cannot be found with a thickness gauge at all. High-temperature hydrogen attack in the synthesis loop, assessed against the API 941 curves, degrades material through the wall without producing measurable loss until very late. Metal dusting in reformed gas coolers and transfer lines produces pitting and localised holes rather than general thinning. Recording the method on every reading is what allows the integrity engineer to see that a component has a long, comfortable thickness trend and no examination at all for the mechanism that will actually take it out of service.

This is why the minimum-per-CML convention is so damaging here. Directional and weld-adjacent attack is invisible in a single minimum. Four quadrant readings trended independently expose a six o'clock rate diverging from the twelve o'clock rate years before the minimum approaches a retirement thickness, which is exactly the lead time you need to buy replacement material for a line that only comes out during a full ammonia stop.

Three arithmetic traps that manufacture corrosion that is not there

The first is coating. Insulated and painted lines in a fertilizer plant carry anything from a thin epoxy to a heavy coal-tar system. A single-echo corrosion gauge measures from the front surface of the coating and reads high; an echo-to-echo measurement between successive backwall reflections ignores the coating and reads true. Change from one contractor to the other and the wall appears to have lost the coating thickness overnight. On a 10 mm line corroding at a genuine 0.1 mm per year, a 0.4 mm mode artefact spread over a four-year gap between campaigns doubles the apparent corrosion rate and halves the remaining life.

The second is velocity. Gauges are commonly left on a carbon-steel default near 5900 m/s. Applied to austenitic stainless around 5740 m/s, the displayed thickness reads roughly 2.8% high, which on a 10 mm urea-grade line is close to 0.3 mm of phantom metal. A plant with mixed metallurgy in the same circuit will collect readings that appear self-consistent within each material and disagree at every transition. Recording the velocity setting used, not merely the material of record, is what allows the error to be found and corrected retrospectively rather than propagating forward.

The third is temperature. Sound velocity in steel falls as temperature rises, so readings taken on a hot stripper bottom without correction read thin, and readings on a refrigerated ammonia tank shell at -33 degC read thick. Neither error is random: each has a known sign, so it accumulates into the trend rather than averaging out. If the surface temperature is not on the row, the correction can never be applied after the fact, and the only honest thing to do with the reading is to distrust it.

Exclusions are data, not gaps

Every campaign in a fertilizer plant leaves readings undone. Insulation was not stripped because the scaffold went up late. A line stayed in service above the couplant's rated temperature. A CML marker was painted over during the last coating campaign and could not be found. The permit for a confined space entry was not issued in the window. In most reporting workflows those CMLs simply do not appear in the returned spreadsheet, and their absence is indistinguishable from a CML that was never on the scope.

The module treats an exclusion as a first-class record with the same weight as a measurement: the CML identifier, the date the attempt was made, the technician who made it, a coded reason, an accountable owner and a date by which it must be closed. Campaign completeness then becomes a number you can put in front of an approver. Signing an inspection as complete when eighteen percent of the scope was never touched is the kind of finding that turns a routine PSM audit into a programme review.

There is a second benefit that shows up a year later. Exclusion reasons cluster. When forty percent of a unit's misses in three consecutive campaigns are insulation-related, the fix is not a better inspection contractor; it is inspection ports, removable covers or a pulsed eddy current screening pass through the cladding. That conclusion is only available if the reasons were coded rather than written as free text in a comments column.

Whose hand held the probe, and were they certified that day

Certification is time-varying, and thickness history is not. A technician certified Level II in ultrasonic testing under a contractor's written practice holds that certification under an employer-based scheme; ISO 9712 certification is issued by a third-party body and behaves differently. Either way it has an issue date, an expiry, an annual vision test and a scope limited to specific methods. The record must hold the certification state as of the date of the reading, not the state as of the day you run the query, because the query you eventually need to run is: show me every reading taken by a technician whose qualification had lapsed.

Operator variation is also real and frequently larger than the signal you are trying to measure. Two competent technicians on the same rough, pitted CML with different couplant, different surface preparation and different gate settings can differ by more than a year of genuine corrosion. Storing the technician on the reading allows that variation to be characterised: when the same person measures the same CML at each campaign, the trend is far cleaner, and where a person changes mid-campaign the discontinuity is at least visible instead of being read as an event.

This is also the field that makes contractor performance measurable rather than anecdotal. Rejection rates at import, exclusion rates by cause, and repeat variance on control CMLs are all computable once the person and the instrument are attached to the number. That turns the next retender into a technical conversation instead of a commercial one.

What the PSM auditor and the API 510 inspector each need

The two consumers of this data want different things and both must be served from the same rows. The PSM auditor wants to see that mechanical integrity inspections and tests were performed on schedule and documented with the five elements 1910.119(j)(4)(iv) prescribes, that deficiencies were corrected before further use or otherwise managed, and that the personnel performing the work were trained for it. That is a completeness and traceability question, answered by the exclusion records, the certification snapshots and the immutable audit trail.

The API 510 or 570 inspector wants a defensible remaining life. That means short-term and long-term corrosion rates computed from named readings, a documented decision about which measurement is treated as the baseline, statistically sound treatment of the reading set, and a clear record of where a fitness-for-service assessment under API 579-1/ASME FFS-1 replaced simple minimum-thickness screening. Both of those questions collapse if the underlying rows carry no method or mode, because then the rate itself is unsourced.

A system that satisfies only one of the two forces a manual reconciliation every audit cycle. A system that satisfies both writes each answer from the same table, and every revision to a reading, an exclusion or a baseline stays in the trail with the reason and the person who made it.

How to evaluate this before you commit

Take a real contractor deliverable from your last campaign, not a sample file, and ask the vendor to import it in front of you. Watch what happens to the rows that are wrong: a duplicated CML, a reading with no instrument, a thickness above nominal, a CML that no longer exists because the spool was replaced. A system that accepts all of it silently is a system that will let a bad campaign into your trend. A system that rejects the whole file is unusable. What you want is row-level rejection with an actionable reason and a queue for the rejects.

Then ask five questions that most demos are not prepared for. Can a reading be stored with a status of attempted but not obtained, and does it appear in campaign completeness? Can an entire contractor campaign be superseded without deleting it, so the original remains visible in the audit trail? Can you show the technician's certification level and expiry as of the reading date rather than today? Can you export the mechanical integrity record with the five OSHA-prescribed fields per reading? And can you find every reading taken with a velocity setting that does not match the component material?

Finally, ask what happens on the day you change contractors. The migration of CML identity, nominal thickness, baseline selection and prior readings into a new firm's field collectors is where most thickness histories quietly lose their provenance. If the answer is a spreadsheet handover, you are about to restart the same problem you are trying to solve. Atlantis provides implementation, ASNT Level III consulting and independent report validation alongside the software; a demo or consultation can be requested at info@atlantisndt.com.

Request a consultation