Turning five contractors' shutdown data into one trustworthy wear trend

Twelve days, eight contractors, one register

A cement plant's inspection year is compressed into a single window. The annual shutdown runs somewhere between ten and twenty-five days, and inside it the kiln alignment and shell scanning specialist, the duct thickness crew, the mill inspection contractor, the refractory surveyor, the electrical and instrumentation teams and often a structural engineer all work in parallel. Every one of them produces data. Almost none of it arrives in the same shape, and much of it arrives after the plant has already restarted.

That timing creates the real pressure. The output the plant needs is not a report; it is next year's work list and this year's purchase requisitions for plate, wear liners and fabricated spools with long lead times. Those decisions are made in the two or three weeks after restart, from data that is still being typed up. Anything that cannot be reconciled in that window gets excluded from the decision, which usually means the reading is filed and the elbow is replaced on the basis of somebody's memory instead.

So the value of a monitoring location register in this industry is not primarily trending. It is ingestion: taking eight submissions in eight formats and turning them into readings attached to locations the plant already recognises, fast enough to matter, with the doubtful ones separated out rather than blended in. A register that makes that reconciliation a two-day task instead of a three-week one changes what the plant can decide before the lead times close.

Abrasive wear is not corrosion, and it breaks the usual rate maths

Nearly all published thickness monitoring practice assumes corrosion: a mechanism that is broadly uniform, roughly steady, and slow enough that a short-term and long-term rate mean something. Cement plant metal loss is dominated by abrasion and erosion from dust-laden gas, plus acid dew point corrosion where flue gas falls below its condensation temperature in a bag filter or downstream duct. Neither is uniform and neither is steady. A riser duct elbow can lose several millimetres at the extrados in a single campaign while the intrados a hand's width away loses nothing.

The consequence is that a wear rate computed from two readings carries almost no predictive weight, because the drivers change annually. A year at reduced clinker production, a change in raw mix chemistry, a higher chloride or sulfur circulation, a period running on alternative fuels, or simply a longer run hour count will all move the loss substantially. The number the register should present is loss per campaign at that exact point, with the run hours and production for the campaign attached, not an annualised rate that implies a physics it does not have.

This also determines how locations are laid out. Where corrosion monitoring can sample a circuit and generalise, wear monitoring cannot. The location must sit at the point of maximum expected loss — the outer radius of the bend, the impingement zone opposite a duct entry, the cyclone cone below the dip tube, the first metre downstream of a damper — and the grid must be dense enough that the worst point falls inside it. A five-point grid on a large elbow will miss the wear scar if it was laid out for corrosion.

Where the location convention actually breaks

The failure is almost never at the gauge. It is at the label. A kiln is scanned in stations measured along the shell, and whether station zero sits at the inlet or the outlet is a convention nobody writes down because everyone on site knows it. The contractor who scanned it last year knew a different one. The result is a dataset that looks perfectly consistent and is mirrored end for end, so the thinning region near the outlet tyre appears to be near the feed end, and the shell plate ordered on the strength of it is the wrong plate.

Clock positions have the same problem with an extra twist: the kiln rotates. A clock reference on a rotating shell is meaningless unless there is a marked physical zero and the record says what it is referenced to — a specific weld, a tyre pad, a painted line at a stated station. Ducts and cyclones fare no better. Elbows numbered in the order a crew walked the elevation will be numbered differently by a crew that started at the other stair.

The fix is boring and completely effective. The plant owns the location identifiers permanently. Each location carries a datum, an offset, a clock or quadrant reference with its physical zero, an access note and a photograph taken from the stance the next technician should use. The contractor receives that list and returns readings against those identifiers only. Everything that arrives without a matching identifier is a question to be resolved by an engineer, not a row to be inserted into the trend.

The gate: what a submission must pass before it becomes a reading

Treat contractor data the way a plant treats incoming material: inspect it before it enters stock. A useful import gate checks, in order, that every row resolves to an existing plant location identifier; that units are declared and consistent; that the measurement mode is stated and appropriate to whether the location is coated; that the instrument and reference block are identified and the instrument was in calibration on the date; that the examiner held the required certification level with a valid expiry on that date; and that the change from the last accepted reading falls inside a physically plausible band.

Rows that fail go to quarantine with the reason attached, not to rejection and not to acceptance. Quarantine matters because most failures are recoverable: a mode field left blank can be confirmed by a phone call, a mismatched identifier is often a transposition, a unit ambiguity resolves as soon as someone looks at the magnitude. What must never happen is a silent decision. Either an engineer resolves the row and it enters the trend with its resolution recorded, or it stays out and the location shows as not read this campaign, which is itself important information.

Keep the contractor's original submission alongside the accepted record, versioned and unedited. When a reading is disputed two campaigns later — and in wear service, it will be — the ability to open the file exactly as it was received, next to the reconciliation an engineer performed on it, is what separates a resolvable question from an argument. It also does something useful commercially: it gives you a factual record of which contractors submit clean data, which is a better basis for the next tender than price alone.

Negative loss and other readings you must never average

Of all the checks, the plausibility band on the delta earns its place fastest, because the negative-loss reading is both extremely common and uniquely damaging. Steel does not thicken. When a duct that lost 6 mm last campaign returns a reading 0.4 mm above its predecessor, something about the measurement changed. Single-echo mode through a coating is the leading cause: paint or a corrosion-resistant coating of a few tenths of a millimetre is included in the reading, and the echo-to-echo technique described in ASTM E797 exists precisely to exclude it. Sitting on a weld cap, a welded wear plate or a doubler produces the same effect at a larger magnitude.

The damage is not the wrong reading, it is what the register does with it. Average a negative delta into a wear rate and the rate falls; a lower rate extends the calculated remaining life; a longer remaining life pushes the next inspection or the replacement out a campaign. The system has produced a confident, documented, entirely false reassurance about the component most likely to fail. Nothing about the output looks suspicious. That is why this must be caught at import rather than reviewed later.

Two related cases deserve the same treatment. A reading far below its predecessor — a 6 mm duct returning 2 mm — is more often a gauge doubling error, a mode-two reading or a point on a different component than a genuine catastrophic loss, and it should trigger a re-measure before it triggers a shutdown. And a location with a first reading well below its stated nominal is usually a baseline problem, since plate arrives inside a mill tolerance and much older ductwork has no recorded original thickness at all. Recording nominal as assumed rather than known is more honest and prevents years of phantom loss.

Contractor qualification is part of the data, not part of procurement

Personnel qualification tends to be treated as a prequalification hurdle: certificates are collected before award, filed, and never looked at again. In an industry where the same crew may not return next year, that is the wrong place for it. What matters is the qualification of the person who took a specific reading, on the date they took it, to the level required by the method. That fact belongs on the reading itself, captured at import and frozen, not inferred later from a folder of certificates with mixed expiry dates.

The same applies to equipment. A thickness gauge's calibration status, its verification against a reference block at the start and end of the shift, and the block's own identity are the difference between a measurement and a number. Where a contractor's own procedure governs — a proprietary kiln shell scanning method, an alignment survey, a wear scanning system — the register should record the procedure reference and revision so that a change in the contractor's method is visible as a possible cause when the trend shifts.

ISO 9001 clause 8.4 already requires a plant to control externally provided processes proportionate to their effect on conformity. A thickness survey that decides whether a preheater duct runs another year is not a low-effect process. Making qualification and calibration mandatory fields at import is the cheapest way to satisfy that requirement, and unlike an annual supplier audit it applies to every reading rather than a sample.

Kiln shells, silos and structures: locations that are not pipe

A large share of a cement plant's monitored population has no pipe-like geometry at all, and a register built only for circuits and vessels will not hold it. A kiln shell is a rotating cylinder monitored by station and clock, where the significant conditions include not only thickness but ovality, axial and radial run-out, tyre migration and creep, and hot spots detected by continuous shell scanning. The natural record is a profile along the shell, and the meaningful comparison is one profile against the last, not a point against a point.

Silos and the preheater tower introduce a different discipline entirely. Reinforced concrete condition assessment uses half-cell potential mapping, cover measurement, delamination sounding, carbonation depth and core results — all referenced to a wall grid whose origin must be fixed to a surveyed benchmark. If the grid origin is re-established by eye at each survey, apparently migrating corrosion zones are an artefact of the grid rather than the structure. The register has to hold the benchmark and the grid definition as part of the location, so that a map produced five years apart is genuinely comparable.

Mill shells and trunnions sit between the two. Their monitoring is crack-based rather than loss-based: magnetic particle or penetrant examination at known stress raisers, weld seams, bolt hole patterns and previous repair areas, with indications recorded by position and length against a marked reference. The register needs to hold indications and repair history with the same seriousness it holds thicknesses, because on a mill shell the indication that was found and monitored is the one that decides whether the plant runs.

How to evaluate a registry on contractor ingestion

Do not evaluate on the dashboard. Take last year's actual submissions — the duct spreadsheet with tabs by elevation, the alignment contractor's PDF, the gauge dump with no mode field, the handwritten sheets — and ask each vendor to load them in front of you. That single exercise separates systems that were built for controlled data entry from systems that were built for the world where data arrives from other people, late, in whatever format the crew's software exports.

Watch specifically for four behaviours. Does an unmatched location code quarantine, or does it create a new location? Is measurement mode a mandatory field with a rule linking it to whether the location is coated? Does a negative or implausible delta stop the row and demand a resolution, or does it flow into the rate calculation? And is the original submission retained, unedited and versioned, next to the accepted values? A system that fails any of these will eventually produce a confident remaining-life figure for a component that is already close to wear-through.

Atlantis NDT builds inspection management software with contractor ingestion as a first-class function: plant-owned permanent location identifiers, a configurable import gate with quarantine and resolution history, mandatory measurement mode and unit declaration, certification and calibration bound to the reading at capture, and profile and grid location types for kiln shells, silos and structures alongside conventional thickness points. It is affordable, accessible and fully customisable. To have it tested against your own last shutdown's files, contact info@atlantisndt.com for a demonstration or a scoped quote.

Request a consultation