Building a pulp mill CML register that survives the next hydroblast

What a turnaround scope list actually has to be

Nobody asks for a CML registry in the abstract. The request arrives about eleven weeks before the mill shut, when someone has to hand a contractor a scope that a bid can be built against, and discovers that last year's scope exists as a spreadsheet with three hundred rows of free text like "lower furnace, left wall, near door". You cannot scaffold from that, you cannot crew from it, and you cannot tell whether the contractor shot the same points as the year before.

A usable scope list is not a list of assets. It is a list of measurement points, each carrying four things: where it is in terms a rigger can act on, what access it depends on, how long it takes to shoot, and what the last three readings were. From those four columns you can derive the two numbers the turnaround manager actually needs — total technician hours and total scaffold square metres — and you can defend them when someone asks why the number went up.

The failure is almost never analytical. Mills know how to compute remaining life. The failure is that the population of points is unstable from year to year, so the scope is rebuilt from memory each cycle and quietly drifts. Once the point set drifts, the trend is no longer a trend, and everything downstream of it is decoration.

The recovery boiler decides the outage, and it erases your markings

In a kraft mill the recovery boiler is both the largest capital asset and the one with the least tolerance for uncertainty. Water leaking into the lower furnace and contacting molten smelt is the explosion scenario the entire BLRBAC recommended practice regime exists to prevent, which is why the annual inspection of pressure parts is treated as a hard commitment rather than a risk-ranked activity. That decision propagates outward: the boiler outage sets the window, and the digester, evaporators, lime kiln and bleach plant all queue behind it.

The consequence for inspection data is specific and physical. The furnace is washed before entry. Hydroblasting at the pressures used for deposit removal will take paint, adhesive labels and shallow marks off a tube wall without difficulty, and chemical cleaning finishes whatever survives. A CML population identified by marks is therefore a population that has to be re-established every single year by a crew standing on a scaffold with a drawing, at the most expensive hour of the mill's calendar.

Tube renewal compounds it. Pad welds, panel sections and full element swaps are routine outcomes of the inspection itself, so the physical thing a CML refers to may not exist next year. The registry has to be able to retire a point at a stated date, open a new one on the replacement, and keep both visible without splicing them into one impossible trend.

Addressing a CML so it survives the wash

The working rule is that a CML address must be reconstructable from a drawing and a tape measure by somebody who has never been in that boiler. In practice that means an ordered set of geometric references: the wall face, the tube number counted from an explicitly named datum, the elevation above the floor tube centreline, and whether the point is on the tube crown or the membrane. Every one of those exists whether or not anyone marked it.

The single most common defect in real registers is an unstated counting direction. One contractor numbers furnace wall tubes from the left facing into the furnace; the next numbers from the left facing out. The result is a mirror image, and a trend that quietly compares tube 14 from one campaign with tube 14 measured from the opposite end. Nothing in the data flags this. The thickness values remain plausible, the corrosion rate becomes noise, and the error is only found when somebody physically walks the numbering and discovers the discrepancy.

The registry's job is to make the datum a mandatory field rather than a convention held in one person's head. Same for elevation reference — floor tube centreline, boiler zero, or grade elevation, each of which differs by a metre or more. Recording which one was used costs nothing at data entry and is unrecoverable afterwards.

Composite tubes: the reading is not the measurement you want

Lower furnace walls in most modern recovery boilers use co-extruded composite tubing, a stainless outer layer bonded to a carbon steel base, chosen because the smelt-side environment will attack plain carbon steel rapidly. This is where a straightforward thickness register goes wrong in a way that is invisible in the numbers.

A standard thickness gauge configured for a homogeneous wall returns the total remaining thickness across both layers. Loss of the protective outer layer produces a small change in that total, so a tube can be well into the degradation that actually matters while the trend line stays comfortably flat. The registry must therefore record the measurand explicitly: total wall, base metal, or clad layer, and the as-new clad thickness for that tube specification. Without those fields, two technicians using different gating produce readings that are not comparable and nobody can tell which is which.

The same discipline covers the cracking mechanisms that composite tubes are known for. Circumferential cracking in the lower furnace is a surface-examination finding, not a thickness finding, so the registry has to hold non-thickness condition observations against the same location identity. If crack indications live in a separate report and thickness lives in the register, no one ever sees that both were found on the same tube.

Digesters, evaporators and the bleach plant obey different rules

The rest of the mill does not behave like the boiler, and a registry that applies one template across all of it will monitor the wrong things. In a batch digester the dominant concern is caustic stress corrosion cracking concentrated in the vapour-liquor interface band, which is why digester inspection guidance emphasises wet fluorescent magnetic particle examination of that zone rather than thickness grids. The location that matters is defined relative to an operating level, and if the mill changes its cooking level, the monitored band has to move with it. A registry that fixes the band at an elevation set in 1998 is watching clean metal.

Multiple-effect evaporators introduce a different problem: mobility. Bundles get pulled, retubed and swapped between effects, and each effect sees different liquor solids, temperature and scaling behaviour. Carrying a bundle's history across a move without recording the change of service produces a corrosion rate averaged over two different duties, which is worse than having no rate at all because it looks authoritative.

The bleach plant is where material-of-construction discipline earns its keep. Chlorine dioxide service pushes you into titanium, duplex stainless and fibre-reinforced plastic, none of which are assessed against a carbon steel retirement thickness and some of which cannot be usefully thickness-gauged at all. The registry needs material and inspection method as first-class fields on the CML, so a point on an FRP tower does not silently inherit a limit that belongs to a steel line.

Corrosion rate is wrong if you divide by calendar time

Pulp and paper is a commodity business with real production curtailment. A mill that took six weeks of market downtime, or ran a digester at reduced rate for a quarter, has an interval between inspections in which the equipment was substantially not in service. Divide metal loss by the calendar interval and you compute a rate that is diluted by idle time, then use it to justify a longer interval, which is precisely the wrong direction.

The correction is unglamorous: store service hours or operating days against each reading interval alongside the dates, and compute rate against both. Where the two diverge by more than a small margin, the running-hours figure is the one that should drive the next interval. Mills that do this typically find the largest divergence exactly where they least want it, on assets that were down because they were already suspect.

There is a second denominator trap specific to boilers. Readings taken before and after chemical cleaning are not comparable, because deposit thickness contributes to the acoustic path and the couplant behaviour on a fouled surface differs from a clean one. Recording surface condition at the time of reading, as a required field, is what lets you throw out the two anomalous campaigns rather than the whole trend.

Sorting scope by scaffold, not by asset

The reason turnaround readiness is the job that sends people looking for this module is that access, not measurement, is the constraint. Shooting a thickness point takes a technician a minute or two. Getting that technician within arm's reach of a tube thirty metres up a furnace wall takes a scaffold crew several days and a permit. So the plan that matters is the access plan, and the registry that helps is the one that can group CMLs by access envelope and report hours and deck area by package.

This changes the conversation with the contractor. Instead of a bid against a list of tags, you tender against a set of packages with known point counts, known technician hours, and a defined completeness criterion. Deferral decisions become visible in advance: if the furnace scaffold cannot be built in time, you know precisely which two hundred points are at risk and what the resulting gap in the trend will be, eleven weeks out rather than on day four of the shut.

It also makes the after-action honest. A package with ninety-one percent completion is a different situation from ninety-one percent of total points shot across the mill, because the missing nine percent in the first case is a contiguous region of the boiler with a known damage mechanism. Completeness by access package is the metric that reflects risk; the mill-wide percentage does not.

Evaluating a registry before the next shut

Ask a vendor to load one recovery boiler wall from your existing records and re-find the points. Do not ask for a demonstration on their sample data. The question is whether your addressing survives their data model, and the answer surfaces within an hour: either the counting datum, elevation reference and clad specification have somewhere to live, or they get flattened into a description field and you have bought a prettier spreadsheet.

Then test the three transitions that break most systems. Retire a point on a tube that was replaced and confirm the trend breaks cleanly rather than continuing across the weld. Move an evaporator bundle between effects and confirm the service change is recorded and the rate resets. Change the monitored vapour-zone band on a digester after an operating level change and confirm the old band is retained as history, not overwritten. A system that handles all three is one you can migrate a decade of history into.

Finally, check what the system produces in the two weeks before the outage without anyone writing a query: a scope list grouped by access package, with point counts, estimated hours, last readings and open recommendations. If that report has to be assembled by hand, the module has not done the job the buyer came for. Atlantis builds this on Odoo so the same record carries the inspection scope, the work order, the contractor and the certification status of the technician who took the reading — affordable, accessible and fully customisable to how your mill already numbers its equipment. Request a demonstration or a scoped quote at info@atlantisndt.com.

Request a consultation