Case: Gradual Drift on an Industrial Compressor
How Rela AI caught progressive degradation on a compressor and coordinated two departments
Case: Gradual Drift on an Industrial Compressor
Context
- Industry: metallurgy
- Equipment: industrial air compressor, 200 HP
- Sensors: discharge temperature, oil pressure, axial vibration
- Criticality: high — feeds the entire plasma-cutting line
The previous problem
The compressor had sporadic failures every 4-6 months. Technicians couldn't predict them because the degradation was gradual — no obvious alarm, just a slow "drift" in temperature and pressure.
- Each failure: 16-24 hours of downtime (diagnosis + waiting for parts).
- Cost per event: ~$25,000 USD (lost production + parts + overtime).
What Rela AI did
Month 1: learning (Level 0-1)
Rela connected to the sensors via OPC UA from a Siemens PLC. During the first month it established baselines:
- Discharge temperature: 85-92 °C (normal).
- Oil pressure: 4.2-4.8 bar (normal).
- Vibration: 2.1-2.8 mm/s (normal).
Day 35: drift detected
The model detected a gradual temperature drift: from an average 90 °C to 94 °C in 5 days. Not an alarm — a trend. AHI dropped from 78 (B) to 65 (C).
Day 38: AI diagnosis
The machine agent analysed the pattern and determined: "Gradual temperature increase with steady oil pressure suggests air-filter degradation or intake-valve wear".
Day 38: multi-department coordination
The agent created TWO work orders with a dependency:
- WO MEC-112: "Inspect air filter and intake valve" → assigned to the mechanical team.
- WO ELE-045: "Verify temperature sensors" → assigned to the electrical team (dependent on MEC-112).
Both technicians received a WhatsApp with full context.
Day 40: coordinated intervention
- Mechanical: air filter 40% blocked + minor valve wear.
- Electrical: sensors correct, confirmed the reading was real.
- Total planned downtime: 3 hours.
Outcome logged
Work orders closed with intervention result: filter replaced, valve cleaned, failure confirmed.
Impact
| Metric | Without Rela | With Rela |
|---|---|---|
| Detection type | Catastrophic failure | Gradual drift caught early |
| Downtime | 16-24 hours | 3 hours (planned) |
| Cost | ~$25,000 | ~$4,000 |
| Cost avoided | — | $21,000 |
| Departments coordinated | Reactive, one at a time | 2 in parallel with dependencies |
| Diagnosis | Trial-and-error | AI suggested the correct probable cause |
What the model learned
- The "temperature drift + steady pressure" pattern got associated with "clogged filter".
- Alert threshold adjusted: trigger when temperature rises more than 3 °C in 7 days.
- The compressor moved up to Level 2 (Predictive) after this first confirmed failure.
This case shows Rela's ability to catch gradual degradation (drift) that doesn't trigger conventional alarms, and to coordinate interventions across multiple departments with dependencies.