One Data Model, Several Rulebooks: Interval Logic Across Marine Sites

Why the numbers do not roll up

A fleet or multi-yard operator asks a reasonable question — how much inspection exposure do we carry, and where is it concentrated — and gets back an answer that cannot be added. One site reports an average remaining life of eleven years. Another reports a list of next survey dates. A third reports percentage of allowable diminution consumed on the critical structural members. None of those is the same quantity, and no amount of spreadsheet work will reconcile them into a single figure that means anything.

The instinctive diagnosis is discipline: the sites are sloppy and need a common template. That diagnosis is wrong often enough to be dangerous. Sites usually differ because they are computing to genuinely different rulebooks. A tanker in class, a bulk carrier on continuous survey machinery, a US inland barge under 46 CFR Subchapter M and a shipyard's own shore pressure plant under state jurisdiction produce due dates by four unrelated mechanisms, with four vocabularies and four definitions of what being overdue means. Forcing them onto one template does not standardise anything. It produces a column of numbers that look comparable, which is worse than an obviously inconsistent report because nobody questions it.

The second reason is subtler and survives even after the rulebooks are reconciled. Remaining life is not an averageable quantity. A fleet's exposure is set by its minimum, not its mean, and the minimum is what determines when a hull comes out of the water. Averaging across a site turns the one asset that binds a drydock date into a rounding contribution. That is not a presentation flaw; it is a structural error in the metric, and it is present in most fleet dashboards that have ever been built.

Class rules are the interval engine — the software's job is to hold several

It is worth being precise about what the software is and is not doing. In marine, the interval engine already exists: it is the classification society's rule set, the flag administration's requirements and, in US inland work, the certificate of inspection cycle. Nobody is asking a piece of enterprise software to invent a survey regime. What they are asking is for the regime to be applied consistently, evidenced properly and made visible across sites that each apply a different one.

That reframing changes the architecture. Instead of a single calculation engine with configurable parameters, you need a set of rule packs — one per regime, each versioned and each carrying an effective date — and a derivation record on every asset that names the pack and the version that produced its current date. When a society revises a rule, you do not silently recompute history. You publish a new pack version, recompute forward, and retain the fact that last cycle's dates were produced under the previous one. Auditors and surveyors ask that question, and a system that cannot answer it has thrown away the evidence for its own outputs.

Continuous survey arrangements make this concrete. Under continuous survey machinery, individual items are credited separately across a five-year cycle, so the vessel does not have one machinery survey date; it has dozens of independently credited item dates whose aggregate defines completeness. A data model with a single "next survey" field per vessel simply cannot represent that, and every site running continuous survey will end up maintaining a shadow spreadsheet — which is exactly the divergence the standardisation project was meant to eliminate.

Coating condition outranks the thickness calculation

In seawater ballast tanks, cargo oil tanks and void spaces, the coating is the protection and the steel is the consequence. IACS condition ratings of GOOD, FAIR and POOR are not descriptive commentary; they drive survey requirements directly, and a POOR rating in a salt water ballast space brings the internal examination forward to annual regardless of what the gauging data says about remaining thickness.

This is a condition switch overriding an arithmetic result, and it is where thickness-only interval engines fail hardest in marine service. Once a coating breaks down, the damage mechanism changes character. General wastage gives way to pitting, grooving at weld toes and edge attack on stiffener flanges — localised, fast, and poorly represented by a general corrosion rate computed from a handful of gauging points. An engine that computes an average wastage rate over the plate and returns a comfortable date is describing a mechanism that has stopped operating.

The data implication is that a condition rating has to be a first-class field with its own history, its own assessor and its own effect on the derived date, not a comment attached to a survey report. It also has to survive the roll-up. Averaging coating ratings across a fleet — treating GOOD as three and POOR as one and taking a mean — destroys the only information the field contained. What a fleet view needs is the count and the location of POOR spaces, because those are the spaces that will set next year's docking plan.

A gauging point is only a point if it is the same point

Thickness trending across drydockings depends entirely on point identity, and point identity is where multi-site marine data most reliably falls apart. The gauging contractor changes between dockings. The reporting format changes. A point that was recorded as frame 62, strake C, 300 mm above the longitudinal is recorded next time as the nearest accessible clean spot, which happens to be 400 millimetres aft and on the other side of a weld seam.

The difference between those two readings is geometry and plate-to-plate variation, not corrosion. It produces rates that are too high, and — just as often — rates that are negative, which the engine will then either clamp to zero or report as infinite remaining life. Both outcomes are wrong in the direction of comfort. Across a fleet, a few percent of points behaving this way is enough to make trend analysis worthless while looking entirely plausible on a chart.

The fix is unglamorous and structural: a stable point identity scheme owned by the operator rather than by the gauging contractor, recorded against the structural drawing rather than against the survey report, and enforced at import. The system should refuse a reading that cannot be matched to a known point, hold it in an exceptions queue, and require a human decision about whether it is a new point or a mislabelled existing one. Refusing data is a feature here. Accepting everything is how a fleet ends up with ten years of history that cannot be trended.

Anniversary windows, credit dates and the drift they cause

Class dates anchor to anniversaries, not to completion. A special survey has a due anniversary and a permitted window around it, and completing the survey early does not move the anniversary forward. The same principle applies to intermediate surveys and to items credited under continuous survey arrangements: the credit attaches to a cycle position, not to the calendar day the surveyor signed.

Almost every general-purpose maintenance system computes the next date as completion date plus interval. Applied to a class regime, that model shifts the cycle every time a survey is done off-anniversary — which is most of the time, because surveys are done when the vessel is available. Two cycles later, the system's dates and the certificate's dates have separated by months, the site starts trusting the certificate and ignoring the system, and the standardisation programme has quietly failed on that vessel.

The requirement is that the engine can hold an anchor date independent of the performance date, express a window rather than a point, and represent credit at item level where the regime works that way. It is not complicated logic. It is simply logic that a system built for plant maintenance does not have, and its absence is the single most common reason marine operators abandon a corporate system and go back to the society's own portal plus a spreadsheet.

Where API 510 and 570 sit alongside class

Marine operators encounter API inspection codes in two places, and confusing them causes real problems. The first is the shipyard's own shore plant: air receivers, steam systems, hydrotest equipment, blast pot vessels. These are pressure vessels under ASME Section VIII with jurisdictional inspection obligations, and where the yard operates a formal owner-user programme, API 510 and 570 supply the in-service interval logic — half of remaining life computed from corrosion rate, bounded by the code's maximum interval.

The second is floating production. An FPSO or floating storage unit carries class on the hull, marine systems and stability, and simultaneously carries process topsides that are inspected to API 510, API 570 and, very commonly, a risk-based programme under API 580 and 581. One asset register, two authorities, two definitions of overdue and two escalation paths. A system that assumes a single regime per vessel cannot represent this at all, and the workaround — splitting the unit into two registers — breaks every roll-up that was supposed to give a single view of the asset.

The design answer is that the regime is a property of the asset, not of the site or the vessel. An asset carries its rule pack, and a vessel is simply a collection of assets that may reference several packs. That sounds obvious written down. It is not how most systems are built, and it is worth checking on the first day of an evaluation rather than the first day of implementation.

Standardising without flattening: what a fleet view should show

If the rules cannot be unified, what does standardisation deliver? Three things. First, one data model: the same asset taxonomy, the same point identity scheme, the same evidence structure, the same definitions of an examination, a finding and a recommendation, so that a person moving between sites reads the same shapes. Second, per-site rule packs that are explicit and versioned, so nobody has to reverse-engineer why a date is what it is. Third, a derived comparison layer that produces metrics which survive the difference in rules.

Two metrics carry across almost every regime. One is the proportion of the allowable diminution or corrosion allowance already consumed on the governing member or component — a unitless number between zero and one that means the same thing on a tanker strake, a barge hull plate and a shore air receiver. The other is months remaining to the earliest binding obligation, whatever rule produced it, reported as a minimum per asset and rolled up as a minimum per site. Neither hides the binding item behind an average.

Everything else — survey names, credit terminology, certificate types — stays local, visible and unmolested. The test of a good roll-up is whether a site superintendent recognises their own numbers in it. If the corporate view has invented a quantity nobody at the site computes, the site will keep its spreadsheet, and the fleet view will describe a fleet that does not exist.

How to evaluate an interval engine for a multi-site marine operator

Bring two of your genuinely different sites to the evaluation, not one. Ask the vendor to configure a class-surveyed vessel with a continuous survey machinery arrangement and a Subchapter M towing vessel in the same system, and then to show you a fleet view containing both. Most products will pass the first half comfortably and fail the second, because the second requires the regime to be a property of the asset rather than a global setting.

Ask which rule pack produced a given date, then ask to see the date as it stood under the previous pack version. Ask what happens to the derived date when a coating rating is downgraded to POOR — it should move, immediately and visibly, without a thickness reading being involved. Ask what the system does with a gauging reading it cannot match to a known point. If the answer is that it creates a new point automatically, you are looking at the mechanism that will destroy your trend data over the next three dockings.

Finally, ask for the anniversary handling in writing. Have them demonstrate a survey completed four months early and show that the next date did not move four months earlier with it. This is a five-minute test that separates products built for marine survey regimes from products built for plant maintenance and relabelled. If you want to run these tests against your own register rather than a demonstration dataset, contact info@atlantisndt.com; consultation and quotes are on request.

Request a consultation