HomeProjectsBearing Early Warning — Vibration Diagnostics with a Live 3D Simulator
Bearing Early Warning — Vibration Diagnostics with a Live 3D Simulator
Early-warning system for rolling-element bearings, built with PT Bara Sakti Ziwa Abadi. It reads raw vibration, compares each bearing against its own healthy baseline, and names the fault from bearing geometry with no training data. Validated on three public run-to-failure datasets and published as an interactive bilingual dashboard with a procedural 3D bearing simulator.
Preview
Highlights
- Reports what it cannot do as plainly as what it can: unproven claims sit next to proven ones
- Smooth 60 fps 3D in the browser: instanced geometry, render-on-demand, 2D fallback for reduced motion
Problem
Industrial bearings usually fail between scheduled inspections, and the standard severity scale (ISO 20816-3) only reacts once damage is already loud. Most predictive-maintenance models also need labelled failure history from the same machine, which a plant installing its first sensor does not have.
Approach
- Envelope-spectrum diagnosis driven by bearing geometry (BPFO, BPFI, BSF, FTF), so it runs on a new machine without a training phase
- Each bearing is judged against its own healthy baseline: amber above 2×, red above 4×, and only once 5 of the last 6 records agree, which is what keeps healthy bearings quiet
- Validated on three public datasets: NASA IMS (12 bearings, 2,063 hours), Ottawa UORED (60 labelled records) and a Paderborn run-to-failure test
- Every rule ported 1:1 from Python to JavaScript and checked against the pipeline's own output, so the dashboard recomputes results rather than displaying screenshots
- Nine-page dashboard in Indonesian and English, with a procedural 3D bearing (no downloaded assets) whose defect pulses at its true frequency, a hands-on fault sandbox, and a live threshold simulator
Stack
Outcome
- Warned before all 4 NASA IMS bearing failures, between 9.6 and 105.9 hours ahead; 2 of 8 healthy bearings flagged, for 0.13% of their running time
- Caught 98% of faulty Ottawa records at 2× baseline, where ISO 20816-3 left 48% of them green
- Paderborn run-to-failure: warned 52 minutes before the end of the test, with false alarms on the healthy section cut from 16.4% to 0% by the 5-of-6 rule
- Named outer-race faults (2 of 2 IMS, 7 of 10 Ottawa). Inner race, rolling element and cage were 0 of 30, and remaining-life hours were never useful on real data; both limits are published on the dashboard
Tags