A single vibration sensor on a critical machine can reveal the early signs of an approaching failure — if the alerting threshold is tuned right. A rolling z-score anomaly detector flags readings that deviate unusually far from recent normal behavior, but ordinary sensor noise also produces occasional spikes.
The Predictive Maintenance Lab monitors one machine's vibration sensor over 200 days, including two simulated approaching failures. Lowering the alert threshold catches the ramp-up to a real failure earlier, at the cost of more false alarms triggered by routine sensor noise.
What this simplified single-sensor view makes clear is a lesson that scales up to far more complex systems: a rolling baseline, rather than a single fixed cutoff, is what lets the same detector work across a machine's whole operating life, even as its normal vibration range naturally drifts over time.
🧪 Try it yourself: the Predictive Maintenance Lab simulation lets you move the alert threshold and watch the 200-day outcome update live.
🧪 Try it yourself: the Predictive Maintenance Lab simulation lets you experiment with everything described above directly in your browser.