Key takeaways
Short answer: Condition monitoring is the measurement of asset state (vibration, temperature, oil quality, etc.) plus alert rules on thresholds. Predictive maintenance uses that condition data combined with failure-history models to forecast when a failure will happen and schedule action before it does. CM is the input; PdM is the output. Plants that confuse them tend to buy CM tools and over-promise PdM outcomes. See also Torque Monitoring vs Cycle Monitoring.
Condition monitoring is the continuous or periodic measurement of asset state. Common types:
CM produces a stream of data. Acting on it requires threshold rules: if vibration exceeds 5 mm/s RMS, alert. CM with rules is widely deployed and produces real value.
PdM uses condition data plus models to forecast:
The forecast comes from models — physics-based, statistical, or machine learning — trained on historical failure data combined with current condition data.
Three reasons:
1. Buying decisions. A CM system is much cheaper than a true PdM platform. If you only need threshold alerts, buying PdM is overspend.
2. Data requirements. PdM needs historical failure data to train models. Greenfield plants without history get CM, not PdM, no matter what the vendor calls it.
3. Expectations. CM tells you when something is wrong. PdM tells you when something will be wrong. Promising PdM with only CM disappoints.
Most plants implement CM and call it PdM because the marketing is better. The threshold rules are condition monitoring; the prediction is implicit (above threshold = will fail soon). This works for many assets but is not the same as a probabilistic forecast with confidence intervals.
The honest position: most "PdM" is really CM with thresholds. That is fine — CM with thresholds delivers real value. Just do not oversell the prediction.
The progression is CM first, then PdM as data accumulates. Plants that skip CM and try to buy PdM directly usually fail because the data foundation is missing.
1. Buying PdM without historical data. Models need history. Greenfield = CM only, until history accumulates.
2. Treating threshold alerts as prediction. A threshold crossing is a current-state signal, not a future-state forecast.
3. Ignoring false-alarm rate. Both CM and PdM produce alerts that turn out to be wrong. Ignored alerts are worse than no alerts. Tune for the false-alarm cost.
4. Not connecting to maintenance workflow. An alert that does not generate a work order rots in a database. Integration with CMMS is non-negotiable.
A modern stack pulls condition data from sensors, applies thresholds for CM-style alerts, and feeds historical failure data plus condition signatures into PdM models for the assets that justify it. The CMMS auto-generates work orders from both CM alerts and PdM forecasts.
Fabrico's CMMS supports CM-style threshold alerts on a wide sensor set and PdM-style forecasting where historical failure data supports model training — both feeding work orders into the maintenance workflow.
See how Fabrico captures this automatically — explore OEE for manufacturing or book a demo.
No. PdM models need condition data as input. CM is the prerequisite.
Depends on failure rate and model type. Years of data is common; for very rare failures, model performance is limited regardless.
Not strictly. Some PdM uses statistical models (Weibull, regression) rather than neural nets. The honest distinction is data-driven prediction vs threshold rules.
Compare false-alarm rate and missed-failure rate against current PM strategy. PdM pays off when it reduces both with cost-effective sensors and modeling.
No. Apply PdM to critical assets where the failure cost is high. Lower-criticality assets get CM with thresholds or run-to-failure.
Login
Blog
