Edge

• Edge versus cloud processing under bandwidth and latency constraints
• Time-series storage, compression and retention for high-rate signals
• Time synchronisation across assets, without which correlation is meaningless

Diagnostics

• Spectral analysis, envelope detection and bearing fault frequencies
• Condition indicators and thresholds grounded in machinery standards
• Anomaly detection where labelled failures are scarce or absent

Prognostics

• Degradation modelling and RUL estimation with calibrated uncertainty
• Run-to-failure data scarcity and using survival analysis with censored records
• Translating a probability of failure into a maintenance decision

Operations

• CMMS integration so predictions become scheduled work orders
• Operator trust, false alarms and the cost of an unnecessary shutdown
• Measuring benefit in avoided downtime rather than model accuracy