Skip to content
Menu

Internal R&D programme

SkAInet: warning of unplanned plant stops months ahead

SkAInet (Smart Knowledge & Artificial Intelligence Network) is our in-house research programme. We are building a machine learning system that detects the slow drift that comes before an unplanned shutdown in a chemical plant, with the aim of giving operators months of warning instead of minutes.

Plant siteYour plantsensors you already haveSkAInetlearns what's normalEarly warningmonths aheadPlanned stopin a quiet week
Illustrative early-warning timelineThe predicted risk of an unplanned stop stays low, then rises slowly over several months. It crosses the alert threshold about three months before the stop, while a conventional alarm only triggers minutes before it.Months of warningStop risk−6−5−4−3−2−10Months before the stopAlert thresholdSkAInet alertConventional alarmUnplanned stop
Illustrative example. The aim is to move the warning from minutes before a stop to months before it.
Plant watch3 models · updated hourlyFailure modeLast 6 monthsStatusPump bearing wearP-101NormalHeat exchanger foulingE-204Driftingplan a stop in ~10 weeksCompressor surgeK-301Normal
Illustrative: each model watches one failure mode.

Predict

Specialised models learn each plant's normal behaviour from the data it already records: temperatures, pressures, flows, motor currents, laboratory results and maintenance logs. Each model watches for one kind of failure.

Explain

Engineers see why an alert was raised. Clear visual comparisons show the current operating state against normal operation and past failures, and a local language model explains the result in engineering terms.

Keep data on site

SkAInet runs on infrastructure inside the plant. Process data, a plant's most sensitive asset, never has to leave the site.

Status: in development. We are targeting chemical and food processing plants in Argentina, the UK and the EU, with the United States on the horizon. We welcome conversations with pilot partners and investors.

Talk to us about SkAInet