RotorVitals — Bearing Fault Diagnosis by Envelope Analysis
Published:
Open, explainable diagnosis of rolling-element bearing faults from a vibration signal — running entirely in the browser, showing its working. Bearings are the most common failure point of crushers, conveyors, pumps and fans; RotorVitals computes the bearing’s kinematic defect frequencies, demodulates the resonance a defect excites, and reads the envelope spectrum for energy at exactly those lines. Live at rotorvitals.fasl-work.com, part of the Faena mining-analytics hub.
The pipeline
signal → band-pass → Hilbert envelope → envelope spectrum → harmonic scoring → diagnosis
- Kinematics — BPFO / BPFI / BSF / FTF from bearing geometry and shaft speed (Randall & Antoni 2011).
- Envelope analysis — band-pass around the structural resonance, the analytic signal via the Hilbert transform, magnitude, then its spectrum. A fault appears as a line at its kinematic frequency.
- Decision — each fault is scored by summed peak energy at its first 5 harmonics, normalized by the noise floor; the top score wins above a floor gate, else “healthy”. Every number traces to a computed line — no black box.
Honest about data
No dataset is shipped or required: the engine is closed-form DSP and the demo signals are generated on-device from published physics (McFadden & Smith 1984), doubling as a self-validation set — the engine recovers the fault frequency that generated each. The SKF 6205/6203 geometries match the open CWRU Bearing Data Center rig, so the same pipeline reads real CWRU recordings unchanged. A single localized defect at steady speed is the easy case; real machines bring variable speed and multiple faults, but the frequency relationships are exact and transfer directly.