Häufig gestellte Fragen
What is the difference between predictive maintenance and preventive maintenance?
Preventive maintenance services equipment at fixed time or operating-hour intervals, regardless of a machine's actual condition, and sometimes replaces components unnecessarily early or too late. Predictive maintenance instead derives the maintenance date from actually measured operating data and forecasts when a failure is imminent. An intermediate stage is condition-based maintenance, which raises an alarm when fixed thresholds are exceeded but, unlike predictive maintenance, does not yet produce a trend forecast of future behaviour. Predictive maintenance thus combines condition measurement with a prediction of the remaining useful life.
Is simple condition monitoring enough, or do you need true predictive maintenance?
For many applications, pure condition monitoring is sufficient: it triggers an alarm when defined thresholds are exceeded and makes acute problems visible. Predictive maintenance goes a step further and uses statistical or AI-supported models to forecast a remaining useful life and a probable failure date from the data history. The additional effort for model building and data maintenance pays off above all where unplanned downtime causes high follow-on costs and lead time for planning offers real added value. For non-critical assets or those that are cheap to replace, simpler threshold monitoring is often the more economical choice.
Which data and sensors are needed for predictive maintenance?
Typical measured variables are oscillation or vibration, temperature, pressure, current draw, acoustics and the quality of lubricants, with vibration analysis considered particularly informative for the early detection of bearing and imbalance damage. These values are captured by sensors at short intervals and passed on via standards such as OPC UA or connections to control and SCADA systems. What matters is not the sheer volume of data but its quality: reliable forecasts require clean measurement and sufficient historical operating and failure data on which the models are trained. Without this data foundation, no robust prediction model can be built.
Which norms and standards are relevant for predictive maintenance?
The general framework for condition-based maintenance is set by ISO 17359 (Condition monitoring and diagnostics of machines - General guidelines), which structures the steps from asset assessment through alarm criteria to prognosis. The ISO 13374 series complements this by describing the data-processing architecture with successive layers from data acquisition through condition detection to prognostic assessment and recommended actions. More specific standards exist for individual techniques, for example vibration measurement on rotating machinery. These standards are not a legal obligation, but they create a shared technical basis and ease the integration of sensor technology, analytics and business systems.
How much can be saved with predictive maintenance?
Industry and consulting studies cite substantial potential, though the actual magnitude depends heavily on the use case. McKinsey puts possible reductions in maintenance costs in an often-quoted range of roughly 10 to 40 percent and also cites a marked reduction in unplanned downtime along with higher asset availability. Such figures are ranges drawn from favourable individual cases, not guaranteed values, since the actual effect depends on the asset base, the data situation and the cost of failures. It is therefore advisable to build your own business case beforehand based on the most expensive unplanned downtimes, rather than adopting blanket percentages.
How long does a predictive maintenance pilot take, and when is it worthwhile?
A pilot project realistically spans around 6 to 12 months, with a substantial share of the time going into data collection and model training, because reliable forecasts need a sufficient data history. It makes most sense to start with a few critical assets where unplanned failures cause particularly high follow-on costs, rather than equipping the entire machine park at once. The most important lever for staying on schedule and achieving quality results is a dedicated internal owner with decision-making authority, plus a clean, sufficiently long data basis. Whether and how quickly the investment pays for itself depends largely on how expensive an avoided downtime actually is in the specific operation.