Industrial predictive maintenance
Know when a bearing is running out of time.
RULGuard estimates Remaining Useful Life from vibration data, and checks whether the model even applies to your signal before it reports a number.
Remaining Useful Life
An estimate, in hours, of how much life a bearing has left based on its current vibration signal.
Health / degradation monitoring
A severity band (healthy, degrading, critical) on the signal over a run — not a specific fault diagnosis.
Applicability checking
Before trusting a number, the system checks whether your signal looks like the data the model was trained on.
Reliability evaluation
Real, measured accuracy of the model against held-out bearings it never trained on.
Supported dataset
FEMTO
Primary model
Extra Trees
Analysis
Recorded acquisitions
Safeguard
Applicability / OOD check
Scope and limitations
- The current model is trained on one public bearing dataset (FEMTO). Other kinds of data are refused, not silently adapted.
- This is offline analysis of recorded data, not a live/real-time deployment.
- No physical fault-type (inner/outer-race, ball, cage) diagnosis is made — only a severity band on the health indicator.