Key takeaways
Only partly, and the mismatch is specific rather than general. Four things have to work differently. The backlog has to be a managed queue, with each job carrying whether it needs a shutdown, which shutdown it is targeted at, and what it costs to defer, because most work cannot be done on demand. Scope building for a turnaround has to be a first-class activity, not a spreadsheet exported once a year. Permits and isolation have to gate the job, since almost everything on a live plant is hot work, confined space or line breaking. And criticality has to assume no redundancy, because a single-train plant has no spare anything.
What transfers unchanged is the measurement. Fabrico calculates availability, performance and quality from PLC data, with IoT sensors and AI cameras where no usable signal exists, and supports conditional tasks and approval workflows that can hold a job behind a permit, an annual maintenance plan, machine registry with documents and history, and downtime, MTTR and MTBF analytics. Setup for the CMMS layer is quoted as 3 days of Fabrico-side work; connecting instrumentation is separate and paced by plant access.
The model most CMMS software assumes is a discrete factory: production runs in batches, lines stop between orders, and a preventive job can be slotted into a gap. A weekly PM calendar makes sense because there are weekly opportunities.
A continuous plant inverts every part of that. Chemical, paper, cement, glass, refining, sugar and many food processes run for weeks or months between planned stops. There is no gap. Work is either done live, which requires permits, isolation and often a bypass that does not exist, or it waits for the shutdown, which may be a year away.
That single fact reshapes everything downstream.
The backlog stops being a symptom and becomes the plan. In a discrete plant a growing backlog is usually a warning. In a continuous plant a well-managed backlog is the shutdown scope forming. What matters is not the size of the backlog but whether every item in it is classified: can this be done live, does it need a shutdown, what happens if it waits another cycle. Our guide to maintenance backlog management covers the general discipline; the continuous-plant variant needs the shutdown field on every item.
Deferral becomes a risk decision, not an admin one. Deciding not to do a job in a discrete plant costs a week. Deciding not to do it in a continuous plant may cost eighteen months, and the failure it prevents will occur at full rate with no bypass. That decision needs to be recorded with its reasoning, because it will be reviewed after any incident.
Inspection is replaced by inference. You cannot open a pump on a running line to look at it. What you have instead is vibration, temperature, current draw, oil analysis and process data. Those readings are the inspection, so they have to be treated with the same seriousness: scheduled, recorded against the asset, trended, and capable of generating work.
A recurring argument says OEE is a discrete-manufacturing metric and does not apply to continuous process. That is half right in a way worth being precise about, because the calculation applies cleanly and only the interpretation changes. The mechanics are in our OEE for manufacturing guide; what follows is what differs.
Availability is not eroded by changeovers, because there are almost none. It is eroded by unplanned trips, and a single trip is expensive out of proportion to its duration because restart is slow. A four minute trip on a continuous line is not four minutes of loss, it is four minutes plus a restart ramp, and if your downtime capture records only the trip you are systematically understating it. Make sure the restart window is attributed to the trip that caused it.
Performance is where continuous plants lose most quietly. Running at 94 percent of design rate produces no alarm, no downtime event and no conversation, and over a year it is usually a larger loss than every trip combined. This is the single strongest argument for machine-level measurement in a process plant: rate loss is invisible to people and obvious to data.
Quality is concentrated in transitions. Startup, shutdown, grade changes and rate changes produce the off-spec, reprocessed or downgraded product. If quality loss is calculated as a flat percentage across the period, the transition cost is smeared into the baseline and nobody ever attacks it. Attribute it to the transition and it becomes a project.
One practical warning on benchmarking. A continuous plant will often show a high OEE simply because it runs continuously, and comparing that number to a discrete plant's is meaningless. Compare against your own history and against design, never across plant types. The same caution applies within a group, as covered in the European multi site rollout playbook.
Question 7 is the one that separates a system that will tell you the truth from one that will flatter you. Ask for a demonstration on real data rather than a yes.
On the maintenance side, Fabrico provides an annual preventive maintenance plan with recurring templates, conditional tasks so a permit or isolation check can gate later steps, approval workflows, calendar and drag-and-drop scheduling for building shutdown scope, a machine registry holding documents and full history per asset, custom fields for criticality and redundancy, and an emergency and reactive flow with reaction time tracking for trips. Inventory covers catalogue, minimum and maximum levels, deliveries, consumption tied to work orders and stock-takes, which matters when a single-train plant carries strategic spares with long lead times.
On the measurement side, availability, performance and quality are calculated from PLC data, with IoT sensors and AI cameras for equipment that exposes no usable signal, a real-time OEE dashboard, micro-stop detection, and downtime, MTTR and MTBF analytics with Excel export. Integration runs through a REST API, webhooks, Excel import and export, and a bidirectional SAP PM sync including S/4HANA, which is the usual landscape in process industries.
Compliance answers, which matter because many continuous plants are also regulated sites: ISO 27001, ISO 9001 and ISO/IEC 20000-1, GDPR data processing agreement, hosting in an AWS EU region, encryption at rest and in transit, daily backups and a 4 hour recovery time and recovery point objective. Support response is contractually under 2 hours.
Two honest scope notes. Fabrico is not a process historian, so continuous process variable trending at high resolution stays in your historian, and Fabrico consumes what is relevant rather than replacing it. And failure prediction is delivered as custom work for specific clients rather than as a standard feature, so if your condition monitoring strategy depends on shipped predictive models, raise it in the first conversation.
Take a plant running a single line at a design rate of 100 tonnes per hour, continuously, with a planned shutdown once a year.
The plant reports availability of 96 percent and considers itself well run. Trips are logged, discussed at the morning meeting, and worked on. Nobody discusses rate, because the line is running and the product is in specification.
Measured from the control system rather than from the shift report, the picture separates. Availability is indeed 96 percent, but the restart ramps after each trip had been recorded as production rather than as loss, so the true figure is lower once ramps are attributed. Performance sits at 93 percent of design, spread across the whole year with no single visible event. Quality shows 1.5 percent off-spec concentrated almost entirely in the hours after each restart and around two grade changes.
The useful conclusion is not the composite number. It is that the largest single recoverable loss was the 7 percent rate gap that generated no event, no alarm and no meeting, and the second largest was a quality loss that had been treated as a fixed cost of doing business because it was averaged across the year instead of attributed to transitions.
Neither was discoverable from the maintenance system alone, and neither was discoverable from people, because both are invisible at the scale a human observes. That is the specific reason machine-level measurement earns its place in a continuous plant, and it is a different reason from the one that applies in a discrete factory, where the losses are visible and the problem is counting them.
It is a useful metric provided you read the components separately and never compare the composite across plant types. Availability, performance and quality each answer a different question and each has a different owner. The composite figure is fine for tracking your own trend and misleading for almost everything else. Some process plants also track total effective equipment performance, which includes all calendar time, because it exposes the cost of long planned shutdowns that OEE excludes by definition.
Give every backlog item two fields: whether it requires a shutdown, and which shutdown it is targeted at. That converts an undifferentiated backlog into a scope document that builds continuously through the year rather than being assembled in a panic. Add the consequence of deferral as a third field and the scope prioritisation argument largely settles itself.
Yes, and more of it is done live than teams assume. Lubrication, condition readings, filter changes on redundant equipment, instrument checks and much inspection can be done on a running plant with the right permits. The discipline is to classify each preventive task by whether it is live-capable, because a plan where everything is assumed to need a shutdown quietly stops being executed.
Include the restart. A trip that lasts eight minutes but takes three hours to bring back to full rate is a three hour event, not an eight minute one. Configure downtime capture so that the ramp period is attributed to the causing event rather than counted as normal production at reduced rate, otherwise every trip is systematically understated and the business case for fixing the cause is understated with it.
Chemicals and petrochemicals, refining, pulp and paper, cement and lime, glass, metals and continuous casting, sugar, dairy and many other food processes, plus water and power generation. Anywhere the plant runs for weeks between planned stops and the product is a flow rather than a count. Related industry guides include OEE software for chemical manufacturing and maintenance software for ATEX hazardous areas, which many of these sites also need.
To see backlog grouped by target shutdown and rate loss reported separately from availability, book a demo and ask for those two views specifically.
Last updated: 7 August 2026.
Login
Blog