Anomaly Detection in SHM

What this topic covers

Identify departures from expected behaviour without assuming every anomaly is damage.

Core monitoring ideas

Anomaly models compare current patterns with historical or modelled behaviour.

Environmental and operational changes can produce legitimate departures.

Engineering review is needed before an anomaly becomes a structural finding.

Interpretation limits

Sampling and storage should match the phenomenon: slow settlement and rapid vibration need different strategies.

Evidence and data quality

Analytics should surface uncertainty, missing data and sensor faults rather than hide them inside a score.

Lifecycle use

AI, machine learning and digital twins support review; they do not replace structural engineering accountability.