Predictive maintenance (PdM)
- Predictive maintenance uses learned models on vibration, electrical and process signals to forecast failures before they stop the line.
- Condition-based programs typically recover 20–50% of unplanned downtime; Renault attributes ~€270M/year in savings to PdM at group scale.
- Two sensing philosophies compete: dedicated vibration sensing (deepest diagnostics on rotating assets) vs. process-signal drift detection (no new hardware, catches non-rotating failure modes).
- PdM does not fix quality or scrap problems — that is process optimization's job, and conflating the two is a common shortlisting error.
Predictive maintenance replaces the calendar with the condition: instead of overhauling a pump every six months, you overhaul it when its signals say the bearing is degrading. The ML layer is what reads those signals — trained on labeled failure libraries (the vibration specialists) or on the asset's own normal behavior (the anomaly-detection approach).
The two sensing philosophies
Dedicated sensing: purpose-built vibration/temperature/magnetic sensors on each asset, feeding models trained on huge labeled fault libraries. Deepest diagnostics — “bearing outer-race fault, medium confidence, act within three weeks” — at a per-asset hardware cost. Process-signal drift: models watch the signals the plant already has (currents, pressures, temperatures, rates) for precursor drift. No new hardware, covers failure modes that never vibrate, and often flags earlier — but with less prescriptive diagnosis. Large plants increasingly run both, and the honest evaluation question is which failure modes dominate your loss history.
What the numbers say
Programs that make it past pilot report 20–50% reductions in unplanned stops, with maintenance-cost reductions following. The headline public case is Renault: about €270M in annual savings attributed to predictive maintenance across the group (2024). The pattern behind successful programs is consistent — start on the assets whose failures actually cost money, wire alerts into the existing maintenance workflow, and measure avoided stops from day one.
FAQ
- What's the difference between preventive and predictive maintenance?
- Preventive maintenance acts on a schedule; predictive maintenance acts on measured condition. Predictive programs cut both categories of waste: failures that arrive before the scheduled overhaul, and healthy equipment overhauled for no reason.
- Do I need vibration sensors for predictive maintenance?
- Only for the failure modes that live in vibration — rotating assets like pumps, fans, gearboxes. A growing share of programs starts from existing process signals instead, which costs nothing in hardware and catches drift-type precursors; many plants combine both.
Every number on this site — in one sourced spreadsheet.
The $43.6B market math, use-case shares, adoption barriers and published deployment results, with the source column intact. Ready to paste straight into your business case.