Anomaly Detection

SUPi continuously monitors incoming data streams — vibration, temperature, pressure, flow — and flags deviations the moment they appear. Not in tomorrow's batch report. Now.

How It Works

The anomaly detection engine uses a multi-layer approach:

1. Statistical baselines — adaptive thresholds that learn normal operating patterns
2. Digital twin comparison — expected vs. actual behavior from physics models
3. Multi-variate correlation — detecting anomalies across related sensor groups
4. Pattern recognition — matching against known failure signatures

Alert Configuration

alert_rule:
  name: "Compressor vibration spike"
  asset_type: centrifugal_compressor
  conditions:
    - sensor: vibration_rms
      operator: ">"
      threshold: "2.5x baseline"
      duration: "5m"
    - sensor: bearing_temperature
      operator: ">"
      threshold: "95°C"
  severity: critical
  actions:
    - notify: [shift_supervisor, maintenance_lead]
    - create_work_order: true
    - log: anomaly_database

Minimizing False Positives

Industrial environments are noisy. SUPi reduces false positives by:

• Cross-referencing multiple sensors before alerting
• Using physics models to validate statistical anomalies
• Learning from operator feedback (confirmed vs. dismissed alerts)
• Applying contextual filters (startup transients, maintenance windows)