Condition-Based Maintenance Layer
Vibration, motor current-signature, and thermal readings are fused into a single condition score per asset, replacing calendar-based PM schedules with triggers based on how the machine is actually running.
The challenge
A heavy-equipment manufacturer maintained its rotating assets — motors, pumps, gearboxes — on a fixed calendar, regardless of actual wear:
- Preventive maintenance intervals were set by OEM defaults, not by how hard a specific asset was actually running
- Some machines were serviced too early, burning technician hours and parts on healthy equipment
- Others failed between scheduled intervals, causing unplanned line stoppages
- Vibration data existed on some assets, current data on others, thermal on a handful — none of it fused into one view
- Maintenance planning had no way to prioritize which of roughly 180 monitored assets actually needed attention this week
How it works
From a calendar to a condition score
The goal was one number per asset that reflected actual health, not time since last service:
- 01
Vibration accelerometers, motor current-signature analysis, and thermal sensors were standardized across all monitored assets
- 02
An anomaly-detection model trained on each asset's own baseline flags deviations across all three signal types
- 03
The three signals are fused into a single condition score per asset, updated continuously
- 04
Maintenance triggers fire when the condition score crosses a threshold, not when the calendar does
- 05
A prioritized work list ranks assets by condition score each morning for the maintenance team
- 06
Technicians log outcomes back into the system, which retrains the anomaly baseline over time
What we built
Key capabilities
One score, three signals
Vibration, current, and thermal readings fuse into a single condition score instead of three separate dashboards to check.
Asset-specific baselines
Each machine is compared to its own normal operating pattern, not a generic OEM threshold.
Prioritized, not scheduled, work
The maintenance team works from a ranked list of what needs attention now, instead of a fixed calendar.
Learns from outcomes
Technician findings feed back into the anomaly baseline, sharpening triggers over time.
Before vs after
Six months of condition-based maintenance
- Maintenance trigger
- Calendar → Condition score
- Unplanned downtime
- 96 hrs/mo → 75 hrs/mo
- Signals monitored per asset
- 1 (inconsistent) → 3 (fused)
- Work prioritization
- Fixed schedule → Daily ranked list
Business impact
What it changed
22% less unplanned downtime
(96 − 75) ÷ 96 hours per month, measured over the first six months of live triggers.
180 assets under continuous watch
Every monitored motor, pump, and gearbox now reports a live condition score instead of waiting for its next scheduled check.
Fewer wasted service calls
Technicians spend less time on healthy equipment and more on assets the data flags as actually degrading.
Technology stack
“Uptime doesn't come from servicing more often — it comes from servicing the right asset at the right time. The plant now knows which one that is, every morning.”
Keep reading