Predictive maintenance
Sensors, models and alerts that flag a developing fault weeks before it stops the line. Rolled out across more than 400 plants.
Unplanned downtime is the most expensive thing that happens in a plant. Predictive maintenance replaces "run until it fails" with "fix it in the next planned window", by watching rotating equipment for the early signs of a fault.
How it works
- Sensors for vibration, temperature and motor current are fitted to the equipment that matters: pumps, fans, compressors, gearboxes, refiners, conveyors.
- Data flows into the same plant historian the control system uses, so there is one source of truth.
- Models trained on that equipment's own behaviour flag changes that match a developing fault.
- Alerts go to the named owner of that asset, with the evidence attached: the trend, the likely cause, the recommended action and the time you probably have.
What we learned from 400 plants
We have rolled this out at over 400 plants and wrote up the three biggest lessons after a 40-site programme for a pulp and paper customer. In short: alerts need a named owner, start with the equipment that has stopped the line before, and models are local to the plant.
Results
| Across monitored equipment | Typical |
|---|---|
| Unplanned stops | Reduced by a third to a half |
| Time from alert to failure | Two to six weeks |
| Payback | Usually the first prevented stop |
Commercial model
Predictive maintenance is sold as a service: sensors and installation, then a monthly fee per monitored asset that covers the platform, model upkeep and a monthly review with a Halden reliability engineer. It is usually bundled into a lifecycle services contract.
Book a discovery call to scope a pilot on your worst five assets.