Case: First Month with No History — Plastics Extruder

Context

• Industry: plastics (injection and extrusion)
• Equipment: twin-screw extruder, line 2
• Sensors: barrel temperature (5 zones), melt pressure, screw speed, motor current
• Challenge: brand-new equipment — ZERO failure history, ZERO prior data

The problem

The plant had just installed a new extruder. No historical data, no known patterns, and the vendor only shipped generic maintenance guidelines every 2,000 hours. The plant manager wanted to know: "When will it actually need maintenance?"

What Rela AI did

Week 1: connection (Level 0)

• Rela connected to the 8 sensors via Modbus TCP from the PLC.
• Maturity Level: Level 0 (Monitoring Only).
• Dashboard showed: "Rela is learning. 3 days of data."

What the operator saw:

EXTRUDER-L2: LEVEL 0 (Detection)
Progress: 30% toward Level 1
Capabilities: ✅ basic alerts | 🔒 health | 🔒 prediction

Week 2: first anomalies

• Rela detected Zone 3 of the barrel swinging ±5 °C (other zones within ±1 °C).
• WhatsApp to operator: "Anomaly on Zone 3 of the barrel. Temperature variation bigger than other zones. Not a failure, but worth watching."
• The operator checked: nothing visible, but took note.

Week 3: Level 1 reached

• With 15 health snapshots, Rela reached Level 1 (Health Tracking).
• Computed AHI: 92 (A) — new equipment, as expected.
• Dashboard now showed degradation trends (flat, as expected on a new machine).

What the operator saw:

EXTRUDER-L2: LEVEL 1 (Health Tracking)
AHI: 92/100 (A - Excellent)
Next level: Prediction (needs 1 logged failure)

Week 4: synthetic data for initial estimate

The plant manager asked: "Can I get a ballpark useful-life estimate?"

• Rela generated a synthetic degradation curve based on the vendor's expected lifespan (10,000 h).
• Initial estimate: "Based on vendor parameters, RUL ~8,500 h. Confidence: 35% (few real data points)."

Month 2: first real failure

• Zone 3 anomaly worsened: ±12 °C swings.
• AHI dropped from 88 to 62 in 10 days.
• Rela alerted: "Significant Zone 3 degradation. Estimated RUL: 200 h."
• A work order was auto-created.

Intervention: Zone 3 heating element with a loose connection. Repair: 1.5 hours.

Record: operator closed the work order and marked "Failure confirmed: electrical".

After the first failure: Level 2

• Rela recalibrated the model with the real failure.
• Maturity Level rose to Level 2 (Predictive).
• The synthetic curve was replaced by real data.
• Confidence rose to 65%.

What the operator saw:

EXTRUDER-L2: LEVEL 2 (Prediction)
AHI: 85/100 (B - Good, post-repair)
RUL: 4,200h (confidence 65%)
Next level: Optimized (needs 2 more failures)

Full timeline

Impact

Bootstrapping lessons

1. Level 0 already has value. The basic alerts caught the Zone 3 anomaly on week 2.
2. Synthetic data is a bridge, not a destination. It gives an initial estimate that gets replaced with real data.
3. The first failure is gold. Every logged failure dramatically improves model accuracy.
4. Transparency builds trust. The operator always knew which level they were in and what was missing for the next.

This case shows Rela AI delivers value from day 1, even without history. The system evolves from basic detection into confident prediction in weeks, not months.