Predictive maintenance · Live
Factory Line A.
Six machines streaming synthetic sensor data. A rolling z-score anomaly detector watches every reading. Click Inject fault on any machine and the model will catch it within ~10 seconds.
Avg health
100
In alert
0
Warnings
0
Tick
0
Sample rate 1.7 Hz · Window 60 samples · Detection threshold z > 1.6
Conveyor 1
100/100
Vibration · mm/s
Temperature · °C
Press 2
100/100
Vibration · mm/s
Temperature · °C
CNC Mill 3
100/100
Vibration · mm/s
Temperature · °C
Pump 4
100/100
Vibration · mm/s
Temperature · °C
Compressor 5
100/100
Vibration · mm/s
Temperature · °C
Drill 6
100/100
Vibration · mm/s
Temperature · °C
Alerts log (0)
All systems within normal operating range. Inject a fault on any machine to see the detector fire.
How the production version differs
- Real sensors via Modbus / OPC UA at 100-1,000 Hz instead of synthetic 0.6 Hz signals
- Per-equipment XGBoost model trained on the client's actual failure history rather than a fixed z-score threshold
- Spectral features (FFT bins, kurtosis, crest factor) on top of the raw vibration trace — much more sensitive to bearing wear
- Edge-deployed on a small Linux box on the factory floor so it works even when the internet is down
- Slack-integrated alerts to the maintenance team so the dashboard lives where they already pay attention