What Five Years of Sensor Data Taught Us About Equipment Degradation Patterns

Five years ago we started logging raw sensor streams from the first 40 industrial assets we had under continuous monitoring. Today that dataset covers 847 assets across 23 facilities — steel mills, automotive press lines, cement plants, chemical process units. We've watched motors fail, predicted them, caught them early, and occasionally missed one. Here's what the data actually shows.

Degradation Is Rarely Linear

The textbook P-F curve is clean. Reality is not. What we see in practice is that most equipment shows no measurable signal degradation for a long time — months, sometimes years — and then the trend bends sharply. A Dodge Dodge-Tigear gearbox on a cement mill raw material conveyor might run perfectly normal vibration levels at 0.15 in/s RMS for 14 months and then hit 0.28 in/s within three weeks. If you're only trending monthly readings, you miss the inflection completely.

Of the 847 assets in our dataset, 71% showed what we call an elbow pattern — stable baseline, then a step change in degradation rate. The average time between elbow detection and actual failure was 23 days. That's your intervention window. It's tight, but it's workable if your monitoring is continuous and your alert thresholds are set correctly.

Bearing Failures Still Dominate

We've catalogued 1,240 confirmed failure events across the dataset. Rolling element bearing faults account for 58% of them. That number has not changed meaningfully year over year. For all the attention given to gearbox failures, motor windings, and hydraulic systems, bearing faults are still what takes equipment down most often in heavy industry.

The good news: bearings give more warning than almost anything else. Our data shows an average of 31 days between first detectable vibration anomaly and functional failure for radial ball bearings in the 40mm–120mm bore range — the size range covering most conveyor head pulleys, fan shafts, and pump impeller shafts. Cylindrical roller bearings on high-speed pinion shafts give you less — closer to 12 days on average. Those need tighter monitoring intervals and lower alert thresholds.

Environmental Factors Matter More Than the Equipment Manuals Admit

We track ambient temperature, humidity, and load cycles alongside vibration and thermal data. The correlation between high ambient temperature and accelerated bearing degradation is real and consistent. Assets in foundry environments where ambient air regularly hits 110°F–130°F show bearing lifecycles roughly 35% shorter than the same make and model in climate-controlled environments at comparable load profiles.

Dust ingestion is worse. Cement plants are the most instructive here. SKF CARB toroidal roller bearings on kiln support rollers — supposed to handle misalignment and contamination — still show significantly faster race wear at facilities with poor dust sealing on housings compared to those that run quarterly purge cycles on bearing housing seals. The bearing is rated for it. The maintenance practice is what varies.

The Failure Mode Shifts With Equipment Age

This one surprised us when we first saw it in the data. For assets under three years old, lubrication-related failures dominate. Over-lubrication causing bearing cage fatigue, under-lubrication causing metal-to-metal contact — both show up clearly in the vibration spectrum as race defect frequencies. Between three and eight years, wear-related failures take over. After eight years, fatigue cracking in gear teeth and bearing races becomes the primary failure mode on gearboxes, and thermal degradation in motor windings on high-cycle motors.

That shift matters for how you set alert parameters. A five-year-old 200kW Siemens induction motor driving a belt conveyor needs different baseline thresholds than a brand-new unit. We now segment our alert models by asset age as a standard practice — something we didn't do in our first two years and paid for with false positives that burned out our customer alert queues.

Time-Based Schedules Miss Both Ends

Before deploying monitoring at a cement plant in the Great Lakes region, the maintenance team was replacing main fan bearings every 8,000 operating hours based on manufacturer recommendation. Our sensor data showed that 40% of those bearings had no measurable degradation at replacement time — they were wasting parts and labor. Meanwhile, 15% had already crossed failure-imminent thresholds before the scheduled replacement. The schedule was both over-maintaining some assets and under-protecting others.

After 18 months on condition-based intervention, they eliminated all 8,000-hour scheduled bearing replacements on fan shafts and replaced them with vibration-triggered work orders. Bearing parts cost dropped 38%. They had one near-miss — a bearing that degraded faster than expected — but caught it with the monitoring system 6 days before it would have failed. Net result: lower cost, fewer surprises.

What the Data Doesn't Tell You

Five years of sensor data is useful, but it doesn't tell you everything. It doesn't capture maintenance quality — whether the new bearing was installed correctly, whether the alignment was checked after installation, whether the housing was cleaned before the replacement went in. We've watched rebuilt assets fail faster than their predecessors because the repair was rushed. The sensor sees the result, not the cause.

It also doesn't capture operator behavior. A stamping press run at 110% rated speed for a production push will show accelerated vibration signatures that look like mechanical wear but are actually load-induced. We've learned to cross-reference sensor data with production logs to separate load events from genuine degradation trends. Without that context, the model generates false urgency.

The Practical Takeaway

If you're setting up condition monitoring on your heaviest assets, here's what five years of data tells us to do: set baseline periods of at least 30 days before activating alerts, segment thresholds by asset age and environment, and monitor bearings on a 10-minute sample interval minimum for high-speed equipment. Don't rely on daily averages for anything turning over 1,200 RPM — you'll miss the early signal entirely.

The assets worth watching most closely aren't the newest or the oldest — they're the ones at the three-to-five year mark where wear rates are accelerating but the maintenance team still thinks of them as "relatively new equipment."