Predictive maintenance that learns a machine from its own sensors

// The problem

A plant's machines fail on their own schedule. Most predictive-maintenance software needs a specialist to set its alarm thresholds, then cries wolf every time a single signal drifts.

// What we built

Software that builds a performance profile of each machine from that machine's own sensor history, compares live readings with the predicted normal and reads the signals together: a hot pipe is not a fault when current and output rose with it. A fuzzy-logic layer names the likely cause.

// Why it matters

Downtime is the most expensive thing in a plant and specialists are scarce. The software was licensed to PTT Exploration and Production.