Key takeaways
Short answer: Sensor fusion combines PLC signals with vision, vibration, audio, and power data to compute OEE more accurately than any single source can. PLC data captures most losses but misses micro-stops, visual quality issues, and degradation invisible to discrete state changes. Adding non-PLC sensors closes that gap, especially on older equipment that was not designed for modern OEE instrumentation. See also OEE vs Utilization.
A PLC reports what its logic was designed to report. Modern PLCs expose run state, cycle counts, fault codes — enough for solid OEE. But three categories of loss are typically PLC-invisible:
These losses are real and PLC-invisible. Sensor fusion catches them.
Vision. Cameras count parts when the PLC does not. Cameras catch visual defects. Cameras detect operator presence and jam-clearing micro-stops. Cheap to deploy, high-value addition.
Vibration. Accelerometers on motors and bearings catch degradation before PLC faults fire. Performance loss often shows up as elevated vibration weeks before MTBF moves.
Audio. Acoustic monitoring picks up cavitation in pumps, bearing degradation, valve hammer. Cheap, retrofittable, often overlooked.
Power. Current and power monitoring on motors detects load anomalies, jams that did not throw a fault, and energy waste during slow cycles.
Temperature. Thermal cameras and discrete sensors catch overheating before failure.
The OEE platform ingests signals from all sources, time-aligns them, and applies rules or models:
Some platforms use simple rules; more sophisticated ones use ML models trained on historical data per asset.
For greenfield plants with modern PLCs, the marginal benefit is smaller. For brownfield retrofits, sensor fusion is often the cheapest path to high-fidelity OEE without changing the PLC.
1. Over-sensoring. Adding vibration, vision, audio, power, and temperature without a clear question for each is a budget sink. Pick the sensors that close specific known gaps.
2. Skipping calibration. Vision systems need lighting consistency. Vibration sensors need proper mounting. Audio needs ambient calibration. Skip these and the data is noise.
3. Treating fusion as automatic. Combining signals requires rules or models. Buying sensors without a fusion strategy gives you data, not OEE improvement.
4. Ignoring data volume. Vision and vibration generate orders of magnitude more data than PLC tags. Storage and bandwidth planning is real.
A modern OEE platform accepts multi-source input — PLC via OPC UA, vision via REST or RTSP, vibration via MQTT or proprietary protocols — and time-aligns them into the unified time-state model. The platform applies rules or models to interpret combined signals as Availability, Performance, or Quality events.
Fabrico's OEE module supports multi-source data ingestion (PLC + vision + vibration + power) with time alignment and rule-based interpretation, plus an extension point for custom ML models per asset — designed for brownfield plants where pure PLC OEE leaves loss invisible.
See how Fabrico captures this automatically — explore OEE for manufacturing or book a demo.
Less so. Modern PLCs expose enough for solid OEE. Sensor fusion most pays off on older equipment or visually-driven processes.
Usually vision (a USB camera or industrial cam plus inference). Catches micro-stops and visual defects at low cost.
Not necessarily. Many fusion rules are simple logic. AI is useful for visual defect detection or vibration anomaly classification.
At sampling rates of 10 kHz+, several MB per hour per sensor. Plan storage and processing accordingly.
Related but not the same. IIoT is the connectivity layer. Sensor fusion is what you do with the data — combining multiple sources into useful signals.
Zaloguj
Blog
