Building the Next Kiln Outage Scope from the Last Outage's Numbers

Nothing thin can be measured while the kiln runs

This is the single constraint that makes cement thickness management different from every process-industry version of the same problem. In a refinery, a piping circuit can be measured on stream almost any week of the year; the inspection interval is a code decision. In a cement plant, the preheater tower's internal surfaces are behind refractory, buried in build-up and at several hundred degrees. The cooler is full of clinker. The kiln shell is too hot for standard couplant and the internal surface is inaccessible until the brick comes out. Measurement is not scheduled by interval; it is scheduled by the plant's stop calendar, and that calendar is set by the market and the kiln's refractory campaign, not by an integrity engineer.

The practical result is that the data set behind the next outage decision is fixed and finite. Whatever was measured in the last stop is what you have. You cannot go back for a clarifying reading, you cannot resolve an ambiguity, and you cannot re-measure a square that returned a suspicious value. Every deficiency in that data set is a deficiency you carry all the way to the day the scaffolding goes up, at which point it becomes an emergent scope item, a change order and an overrun.

That is why the provenance fields matter here for a different reason than they do in a pressure-equipment plant. There, provenance defends an inspection interval. Here, provenance is what lets you decide, months in advance and without the ability to check, whether a number you are looking at is real.

Cement wear is not code corrosion

The mechanisms are mechanical and thermal far more than they are electrochemical. Clinker dust in a high-velocity gas stream abrades the outer radius of every elbow and the entry region of every cyclone. Alkali and sulphur circulate through the kiln and preheater, forming build-up that both protects and, when it spalls, tears at the steel underneath. Cold-end equipment, particularly when the raw mill is down and the gas is not being dried, drops below acid dew point and corrodes in a pattern that tracks cold-running hours rather than clinker tonnes. The kiln shell itself sees thermal cycling, ovality-driven flexing and local overheating wherever refractory has failed.

None of these produce the smooth, near-linear wall loss that a corrosion rate is designed to describe. A duct elbow can hold nominal for two campaigns and then lose four millimetres in one, because a build-up shelf collapsed and exposed bare steel to the stream. Trending such a component with a single linear rate is not conservative or unconservative; it is simply the wrong model. What the history has to support instead is per-square rates, visible acceleration, and a comparison of this campaign's map against the last one so a new scour pattern is obvious at a glance.

There is also no external authority setting the rules. Cement and lime plate equipment is largely outside the scope of pressure-equipment codes; there is no API 510 equivalent for a cyclone cone and no statutory inspection interval. Sites therefore adopt internal retirement criteria, often informed by API 579-1/ASME FFS-1 methods applied voluntarily, and those criteria live or die on the consistency of the underlying record. When the criterion is a site standard rather than a code, the record has to be more disciplined, not less, because there is no external body forcing the discipline.

The grid has to land on the same steel twice

A wear map is only a trend if the second campaign measured the same physical squares as the first. In practice, grid registration fails constantly. Paint markers are blasted off during the outage coating campaign. The scaffold is built differently and the crew starts numbering from a different corner. A cone is described as having twelve segments in one report and eight in the next because the reporting crew counted plate seams instead of design segments. The result is two beautiful data sets that cannot be subtracted from each other.

The fix is to define the grid origin against something permanent and physically findable: a manway centreline, a specific weld seam intersection, a nozzle, a permanently stamped datum punch. The record then holds the origin, the pitch, the axis convention and the numbering direction, and the job pack issued to the next crew carries the previous values pre-loaded square by square. A technician who can see that square C-7 read 8.2 mm last time and reads 4.1 mm now will stop and check rather than write it down and move on.

Pre-loading prior values has a second effect that surprises people. It changes the incentive structure of the field work. A crew collecting numbers into an empty template has no way to notice an error; a crew collecting against a prior map is effectively performing a continuous plausibility check, and the rate of transcription errors, misregistered grids and swapped axis conventions falls sharply. That improvement costs nothing except the ability to issue the history back out as a job pack, which is a capability worth testing in any demo.

The plate-replacement discontinuity

This is the failure mode that quietly destroys cement thickness histories, and it has no equivalent in a refinery piping circuit. Outage work replaces steel. A cone segment is cut out and a new one welded in. A worn duct panel is plated over with a doubler. A wear liner in abrasion-resistant plate is fitted where none existed. Each of these events changes the material, the nominal thickness and the retirement criterion at that location, and each of them breaks the arithmetic of any trend that runs through it.

Three specific errors follow. Replacing worn 6 mm steel with new 12 mm plate produces a negative wear rate at the next campaign; systems that clamp negatives to zero make the location look stable when it has actually been reset. Replacing carbon steel with a thinner abrasion-resistant grade produces an apparent step loss that triggers investigation into damage that never occurred. And a doubler plate welded over a thinned panel produces a UT reading through two plies with an air gap, which reads either as the outer ply alone or as an unstable echo, depending on fit-up.

The record therefore needs component-level events, not just readings: an installation event carrying date, material grade, nominal thickness, extent and the work order that produced it. The prior grid retires against the removed steel and remains available for historical analysis; a new baseline starts from the installation date. Retiring history rather than deleting it is the point. Ten years later, when you want to know how long the last cone lasted before replacement, the retired grid is the only place that answer lives.

Exclusions inside a fourteen-day window

A cement outage is a queueing problem in which inspection competes with refractory, mechanical and electrical work for the same access. Inspection is usually near the back of that queue because the areas it needs are the areas everyone else is working in. The consequence is that a meaningful fraction of the planned grid never gets measured, and the reasons are almost always about access and sequencing rather than about the steel.

Coded exclusions turn that into a manageable pattern. When the same eight squares on the same cyclone are skipped for build-up in three consecutive stops, the conclusion is not that the crew was lazy; it is that the cleaning sequence needs to put that cone ahead of the inspection window, or that a permanent access platform is justified. When the misses cluster around scaffold availability at a particular elevation, that is a planning input for the next outage's scaffold schedule, and it is worth real money because scaffold is one of the largest single cost lines in a stop.

Exclusions also protect the scoping decision itself. A component whose grid was eighty percent measured supports a confident replacement decision. One whose grid was thirty percent measured, with the missing squares concentrated in the region that historically wears fastest, supports no decision at all, and the honest output is a contingency plan and a pre-ordered plate rather than a false sense of coverage. That distinction is only visible if the misses were recorded.

Two methods, two quantities, one dangerous chart

Cement plants increasingly screen insulated ducting with pulsed eddy current, because it works through cladding and does not require lagging removal or full scaffolding. It is a genuinely valuable technique and it belongs in the programme. What it does not do is produce a number that can share an axis with a contact UT reading. Pulsed eddy current, applied per ISO 20669, reports an average wall thickness over the probe footprint, referenced to a calibration on nominal material. Contact UT per ASTM E797 reports the thickness under a probe a few millimetres across.

On uniformly corroded plate the two agree reasonably well. On the scoured, patchy loss that cement ducting actually suffers, the PEC average sits well above the UT minimum, and by a margin that depends on how localised the scour is, which is exactly what you do not know in advance. Plot a PEC value from 2023 and a UT value from 2026 on the same trend line and the chart shows a loss event that is entirely an artefact of the method change. The reverse ordering shows reassuring growth.

The discipline is simple to state and easy to enforce once the method is on the reading row: series are per method, screening results drive follow-up rather than trending, and a component's retirement decision is taken on the measurement type its criterion was written for. Any system that lets you build a single trend across mixed methods without a warning is a system that will eventually be used to make a wrong call about a cyclone.

From history to a scope list, a steel order and a crew plan

The output that justifies the whole exercise is not a thickness report. It is a ranked scope list that answers a question with a hard deadline: which components will reach their replacement criterion before the outage after next, given the wear rate measured on their own grid and the plant's confirmed stop calendar. That framing forces two things most inspection reporting omits. It forces a date for the next window and the one after it, and it forces the wear rate to be expressed per square with a confidence that reflects how much of the grid was actually measured.

From that list falls everything else. A materials take-off with grade, thickness, extent and lead time, driving purchase orders raised months ahead of the stop. A labour estimate in welder-days and scaffold square-metres. A decision on whether a component is plated, part-replaced or fully replaced, which is often the difference between four days and eleven days of critical path. And a residual list of components that were measured too thinly to decide, which is the honest input to the next campaign's inspection scope.

The loop closes when this outage's plan becomes next campaign's job packs, pre-loaded with prior values and issued to whichever crew is on site. Atlantis implements this as a working cycle rather than a reporting module, alongside ASNT Level III consulting and independent report validation. Affordable, accessible and fully customisable to your component hierarchy and site conventions; a demo or consultation can be requested at info@atlantisndt.com.

Request a consultation