Glossary · IIoT, data and AI
Anomaly detection
Also known as: Outlier detection, Novelty detection
German: Anomalieerkennung
In industrial data analytics, anomaly detection is the identification of data points, patterns or time periods that deviate significantly from expected or learned normal behavior. Methods range from fixed limits and statistical tests to unsupervised machine learning models.
- IIoT
- AI
In one sentence
Anomaly detection flags machine or process data that deviates from normal behavior, using limits, statistics or machine learning.
Example
An unsupervised model trained on six weeks of spindle vibration data flags an unusual frequency pattern two days before a bearing is replaced.
How it applies
- Engineering: Define what normal means: operating modes, product types and seasonal effects all change the baseline. A model trained on one product can flag every run of another as anomalous.
- Operation: An anomaly is a hint, not a diagnosis. Route flags to people who can interpret them, and avoid adding them to the alarm system without alarm rationalization.
- Maintenance: Combined with Condition monitoring, anomaly scores help prioritize inspections; confirmed findings improve the model over time.
- Documentation: Describe the input signals, the training period, the threshold logic and what operators should do when a flag appears. Record retraining as a model change (Model versioning).
Anomaly detection vs. alarm
An alarm in the sense of alarm management is an engineered, rationalized notification that requires a timely operator response. An anomaly flag is a statistical observation that may or may not matter. Treating anomaly flags as alarms without that rationalization quickly leads to alarm floods.