Predictive Monitoring Pays Back at Scale-Up

A refinery reliability team, referenced in a Plant Engineering feature published on 18 August 2026, instruments about 10% of its rotating equipment against known bad actors, watches a lubrication issue surface within minutes on a vibration sensor that had only just been activated, and lands a proof of concept. Every executive in the room agrees to a wider rollout. Eighteen months later the programme sits at 14% instrumented, and the maintenance report still shows the same availability figures as before the pilot. That is the harder half of the industrial predictive monitoring story, and it is where the vendor briefings end.

Where the Plant Engineering Argument Is Right

The 18 August piece, “Drive up plant production, quality with asset management programs”, is worth reading. It sets out a criticality-tiered monitoring model: monthly manual routes at low criticality, hourly wireless at medium, continuous once-per-minute at high, and sub-100-millisecond monitoring on assets carrying catastrophic risk. It names the failure mode where an IT-picked, low-capability sensor delivers raw counts and no context. It puts Category 3 and 4 vibration analysts in the loop, and it warns against instrumenting the plant’s best assets to demonstrate ROI. For an operator who has not yet started, that model is a serviceable brief.

Where the Argument Stops Short

The Plant Engineering piece treats the pilot as the point of decision. In an operating Maximo or MAS estate the pilot is the easy part. The difficult work sits at scale-up, when the programme moves from 10% of instrumented assets to 30% or 40%, and three preconditions in the CMMS decide whether the availability numbers actually move. All three sit with the head of asset management. The vendor stack supplies sensors, models and dashboards; the preconditions sit outside its contract.

Precondition One: Criticality a Director Can Defend at Audit

ISO 55001 is explicit that lifecycle decisions must be traceable to a documented risk and criticality assessment. Auditors look for capital plans and maintenance strategies that reference the same criticality tier the operating plan uses. In most estates the criticality field on the asset record was set once, in a workshop six years ago, from a spreadsheet nobody has since opened.

At pilot scope that gap is invisible: the 10% is chosen from known bad actors, and the tier debate is deferred. At scale-up the same gap decides which pump gets the vibration sensor and which one waits another quarter. Without a criticality assessment methodology applied at commissioning, reviewed after any material change in consequence, and owned by a named role, the tiering the article recommends collapses into whichever plant manager has the loudest voice at the monthly review. Where a register does not carry this today, criticality frameworks that survive an audit are the place to start.

Precondition Two: Failure Coding That Supports a Threshold

An intelligent vibration sensor that reports “low lubrication” against a rotating asset is only useful in the CMMS if the receiving system knows which job plan and which craft to raise. That decision depends on failure mode classification, which for asset-intensive operators means an ISO 14224 taxonomy applied consistently against every closed work order for at least twelve months.

Most CMMS estates do not carry that. Free-text closures, mis-classified failure codes and inconsistent hierarchies are the ambient state. An anomaly output that lands against an asset whose historical failure record is unstructured becomes a service request the shift supervisor cannot triage. The work goes into a specialist queue, the specialist queue lengthens, and the availability figure holds where it started. This is a data programme before it is a monitoring programme, and it is why failure coding disciplined enough to earn its keep is a precondition, not a follow-on.

Precondition Three: Sensor Work Sits First-Class on the Schedule

The third failure mode is scheduling. In most Maximo estates, sensor-triggered service requests arrive on a separate intake, land in a supervisor’s review queue, and are actioned only when planned work already leaves headroom. Planned work rarely leaves headroom, so the sensor-generated work waits, and by the following month it has been deferred into the next review.

Plant Engineering names the workforce constraint. The scheduling constraint sits alongside it. Continuous monitoring shortens downtime only if the sensor’s finding lands on the same weekly schedule as PM work, with the same job plan library, the same craft allocation and the same acceptance criteria. That means a documented SR-to-WO route with a service level attached, a scheduler who treats sensor findings as scheduled work rather than as an exception, and a supervisor who closes the loop against the failure code.

The Coverage the Availability Figure Actually Rewards

The Siemens True Cost of Downtime 2024 report puts the aggregate loss across Fortune Global 500 manufacturers at roughly USD 1.4 trillion a year, with an idle automotive line at USD 2.3 million per hour and consumer packaged goods at around USD 36,000 per hour. Those figures are widely cited but illustrative, and they are the numbers vendors quote at the pilot decision.

The number that moves the reliability report is coverage over time. The MaintainX 2026 State of Industrial Maintenance survey found 58% of maintenance teams already using AI, against 79% reporting unplanned downtime unchanged or worse, and 39% saying downtime events are now more expensive than a year ago. Coverage that stalls at 10% is what closes that gap in the wrong direction. Coverage that reaches 30% to 40% of criticality-classified rotating assets, under the three preconditions above, is where the aggregate downtime number starts to move on the monthly report rather than in the pilot deck. The distance between those two points is a governance gap.

What Stays Hard

Two things stay hard even where the preconditions are met. Category 3 and 4 analyst time remains scarce, and building an internal bench takes several planning cycles. The vendor-partner path the Plant Engineering piece describes is a reasonable answer, provided the partner is contracted against measurable coverage growth per quarter rather than against a sensor count on a purchase order.

Retrospective data quality is the second. Most operators will not recover the full historical archive of free-text closures. A pragmatic head of asset management sets a cut-off date, commits to ISO 14224 coding from that date forward, uses the following twenty-four months as the analytics baseline, and states in the capital paper that the model input begins at the cut-off. That is a defensible position with an auditor and with a CFO.

Position

The Plant Engineering piece is right that asset management is now a board-level concern and that criticality-tiered monitoring is the operating shape for the next planning cycle. It stops one step short. The pilot is where the technology proves itself. Documented criticality, ISO 14224 failure coding, and a first-class SR-to-WO route are what decide whether coverage moves past pilot scope, and whether the availability figure follows. Operators who go into their next capital paper with those three lines named as preconditions will spend the sensor budget once. Operators who leave them to the integrator will run the pilot again in eighteen months, with the same executives in the room and a different vendor logo on the slide.