Making six mills' thickness numbers add up to one number

The assets do not resemble each other, and neither do the histories

A corporate integrity function looking across six mills is not looking at six copies of the same programme. The recovery boiler is the defining asset: a Section I pressure vessel firing black liquor, where a tube leak into the smelt bed is a smelt-water explosion, and where inspection practice is shaped by BLRBAC recommended good practice as much as by code. Lower furnace waterwall and floor tube thickness, primary air port openings and smelt spout openings are inspected every outage, and the decisions are made under a hazard that has no analogue in most industries.

The digester is a different discipline entirely. In a continuous Kraft digester the dominant concern is alkaline stress corrosion cracking in weld heat affected zones, found by visual and surface methods, with wall loss a secondary and slower issue. A thickness programme built around the digester will be organised around cracking findings and weld overlay history, and a thickness table imported from it will have a structure nothing like the recovery boiler's.

The bleach plant breaks the pattern again. Chlorine dioxide stages run acidic chloride chemistry that eats conventional stainless, so the equipment is titanium, high-alloy or fibre-reinforced plastic, and much of it cannot be ultrasonically trended at all. Then there are the evaporators, the lime kiln shell, green and black liquor lines with erosion at bends and the ClO2 generator. Six mills times this asset spread is why nobody in the group can answer a simple question about fleet condition.

On a composite tube, what exactly is the thickness?

This is the sharpest example of the standardisation problem because it is not about administration, it is about physics. Recovery boiler lower furnace tubes are typically co-extruded, with a stainless layer metallurgically bonded over a carbon steel base tube. Ultrasonically that bond line is an acoustic interface and it returns an echo. Depending on gate setting, gain, probe and instrument, the reported number can be base metal remaining, total wall including cladding, or something that has locked onto the wrong echo entirely.

Two mills using the same instrument model and the same nominal procedure can therefore be recording two different quantities under the same column heading. The difference is roughly the cladding thickness, and it is easily larger than a decade of wall loss. Worse, when a mill changes contractor and the new crew sets up differently, the step change appears in the trend as sudden thinning or sudden thickening on a component where those readings drive tube replacement decisions.

The record has to carry the measurement definition, not just the value. Which layer the reading represents, the setup used, the calibration reference and the technique document revision. Then a roll-up can compare like with like, and a step change between contractors can be diagnosed as a technique change in an afternoon rather than argued about across an outage. No amount of governance policy substitutes for that field existing.

Baseline: nominal, measured, or the thinnest thing anyone ever saw

Ask six mills what the starting thickness of a line was and you will get three answers. One takes nominal from the line list. One takes the first measured reading. One takes the lowest reading ever recorded, on the reasoning that it is conservative. Each is defensible in isolation and they are not comparable in aggregate, which is the whole problem.

The arithmetic matters. Seamless pipe to ASTM A106 or A53 is permitted twelve and a half percent undertolerance on nominal wall, so a nominal 0.280 inch line may legitimately leave the mill at 0.245 inch. A site baselining on nominal records a phantom 0.035 inch of loss in the first inspection interval, computes a spectacular corrosion rate, then watches it collapse toward the true rate over subsequent cycles. A site baselining on first measured never sees that artefact. Rolled up together, the nominal-baselined mill appears to have the worst corrosion in the group and gets the capital.

The resolution is to store both, and to store which one each derived number used. Nominal is a property of the specification. The measured baseline is a property of the measurement record, with its own date, technician and instrument. The corporate rate is then computed from the measured baseline where one exists and flagged as nominal-derived where one does not, so the reader can see which mills have a real baseline and which are still guessing.

Long-term, short-term, and the rate that ranks mills by arithmetic

API 570 draws a useful distinction that mills adopt inconsistently. The long-term rate is computed from the initial thickness and the current thickness across the full elapsed period. The short-term rate uses the previous reading and the current one. The code practice is to compute both and use whichever produces the shorter remaining life. In multi-mill reality, one site uses long-term because it is smoother, another uses short-term because it is conservative, and a third uses a regression across all readings because an engineer once built the spreadsheet that way.

On a component whose corrosion has recently accelerated, these produce materially different answers from identical data. A regression across fifteen years drags the recent slope back toward the historical average and reports comfortable remaining life on a line that has started thinning three times faster since a process change. Short-term catches it and looks alarmist on components that are genuinely stable but have measurement noise. Neither is wrong. Choosing per mill and then adding the results is wrong.

The design conclusion is that a corporate system should store readings, not rates. Rates are derived, and derivation belongs at the centre, applied uniformly, recomputable on demand. That way a decision to change convention is a configuration change that re-derives the whole fleet consistently, rather than a memo asking six mills to rebuild spreadsheets they will rebuild slightly differently.

Units, rounding and the silent factor of 25.4

The units problem sounds too trivial to cause harm and it causes harm constantly, because units in most inspection data live in a column header rather than on the value. A mill switching from an imperial to a metric contractor, or importing a data set assembled by a corporate intern, can end up with a column containing both. Nothing about the number tells you which it is, and a plausible-looking imperial thickness and a plausible-looking metric thickness are separated by a factor most reviewers will not notice at row level.

Rates are worse. A corrosion rate quoted as five is five mils per year at one mill and five millimetres per year at another, and those describe a component that will outlive you and a component that will not last the year. Rounding compounds it: a site reporting to the nearest mil cannot resolve a real loss of half a mil per year, so its components trend flat until they suddenly do not, while a site recording to two decimal millimetres shows the slow loss clearly and appears to have a worse problem.

The rule is simple and rarely implemented. Store the unit and the precision with each value, not on the column. Store the raw as recorded, then convert for display and computation, never in place. And make the roll-up refuse to aggregate values whose unit is unknown rather than assuming a default, because an assumed default is precisely how a factor of twenty-five enters a board report.

Exclusions and carry-forward: the healthiest points are the ones nobody reached

Mill outages are short, contested and weather-dependent. Points get missed. A digester platform was not available, the recovery boiler floor was still too hot on the day the UT crew was scheduled, the insulation contractor did not get to the black liquor line, the lime kiln was still turning. All ordinary. What varies between mills, catastrophically for roll-up purposes, is what happens to the cell.

One mill leaves it blank. One writes N/A. One carries forward the previous reading so the sheet looks complete, which is the most damaging option because it produces a defensible-looking zero corrosion rate at exactly the locations nobody has been able to inspect. Aggregate across six mills and the resulting fleet picture systematically understates risk in inaccessible areas, which is the same population as the areas where access is hard because the geometry is awkward, and awkward geometry is where erosion concentrates.

A common exclusion taxonomy is one of the few things worth genuinely mandating. A short governed list of reason codes, mapped from each site's historical local usage, with an authorising name and an access requirement attached. Then two reports become possible that no mill can produce alone: true coverage against planned scope across the fleet, and a ranked list of points excluded for multiple consecutive cycles. The second is usually the most valuable inspection scope document the group owns.

One canonical identifier, many local names

The instinct when standardising is to impose a numbering scheme and renumber everything. It fails. Field crews work from the tag painted on the pipe and the sketch in the binder, the mill has ten years of work orders referencing local names, and a renumbering exercise breaks the join between history and asset silently, usually at the point of a revamp when a section was replaced and the numbers were reused.

The pattern that works is a canonical corporate identifier that is never displayed to a field crew, plus an alias table holding every local name each point has ever had, with the date range each was valid. Readings bind to the canonical key. Field interfaces, printed rounds and contractor deliverables all work in the local name. Import maps the alias, and where a name is ambiguous, the import fails rather than guessing, which is the whole point.

This also solves the succession problem. Aliases capture the 2016 renumbering, the contractor's own scheme, the drawing tag and the CMMS functional location. When someone asks in five years why the trend at a point breaks in a particular quarter, the alias history answers it in seconds. Without it, the answer is a retired engineer's memory, and eventually the answer is that nobody knows and the history before that date is discarded.

What corporate should ask a vendor to demonstrate

Bring the two most different mills you own to the evaluation, not the tidiest one. Ask the vendor to load both data sets and produce a single ranked list of worst components across them, with every derived number labelled by the convention used to derive it. The interesting moment is not whether the list appears. It is whether the system tells you that one mill's baselines are nominal and the other's are measured, or whether it silently ranks them against each other anyway.

Then test the recomputation. Ask them to change the corporate corrosion rate convention from long-term to the shorter of long-term and short-term, and to re-derive the whole fleet without touching a single stored reading. If that requires data migration, rates are being stored rather than derived and every future convention change will be a project. Ask the same of the governing minimum, since fitness-for-service assessments will keep changing it on individual components.

Finally, ask to see an exclusion, a carry-forward rejection and an alias collision. A system that accepts a workbook with a repeated value at an unreachable point, no unit on the column and an ambiguous local tag is not going to standardise anything. It will simply produce a cleaner-looking version of the disagreement you already have, which is a worse position than the spreadsheets, because the spreadsheets at least look like they need checking.

Request a consultation