Anomaly detection
also: outlier detection · fault detection
Automatically flagging data that departs from normal behavior, such as unusual vibration or energy use, to catch faults before they become failures.
Instead of defining every possible fault in advance, anomaly detection learns what "normal" looks like from historical data and flags deviations. Methods range from simple thresholds and statistical limits, through models of expected behavior (residuals from a physics model or digital twin), to machine-learning approaches. Applications include equipment monitoring, energy-waste detection, network security and production test data.
The hard parts are practical: defining normal across operating modes (startup, seasons, different loads), keeping false alarms low enough that people don't learn to ignore them, and turning a flag into an actionable diagnosis.
Founder takeaway
Start with simple, explainable methods on well-labeled data, and measure false-alarm rates with the people who will respond to the alerts.
engineer review pending