Abstract
Predictive maintenance based purely on statistical anomalies is insufficient for critical infrastructure. True Equipment Health Intelligence maps raw telemetry directly to known physical failure mechanisms to prevent unscheduled downtime.
Condition-Based Monitoring (CBM) Constraints
Traditional CBM relies on static alarm thresholds (e.g., triggering an alert if vibration exceeds 4.5 mm/s). Anomaly detection via generic machine learning improves signal timing but fails to provide root-cause diagnostics, leaving plant operators with an alert but no actionable engineering context.
Decomposing the Reliability Signature
A centrifugal chiller's health is a composite of distinct physics. Mechanical degradation is isolated via spectral vibration analysis (monitoring 1X RPM for imbalance, or high-frequency blade pass frequencies for aerodynamic stall). Thermodynamic health is tracked via approach temperatures and sub-cooling variance. Electrical health is monitored via motor winding resistance. Each requires specific sensor architectures.
Fault Tree Analysis
Engineering intelligence platforms must execute automated Fault Tree Analysis. If a system detects a rising condenser approach, it must cross-reference condenser flow and compressor lift to diagnose tube fouling versus a failing pump. The output must be an explicit diagnostic trail linked directly to maintenance actions.