Also called condition-based maintenance, CBM, condition monitoring
What is predictive maintenance?
Servicing equipment when its measurements say it is heading for failure, rather than on a schedule.
What else it is called
- Other names
- condition-based maintenance, CBM, condition monitoring
- In French
- maintenance prédictive
Related entries
Understanding predictive maintenance
Rather than servicing at a fixed interval, the equipment is monitored and work is triggered when the measurements drift away from their normal pattern. A bearing develops a vibration signature before it seizes, a motor draws more current as it struggles, oil carries metal before a gearbox fails. Catching those gives you a planned stop instead of a breakdown, and uses the full life of the component.
This is the application vendors mean when they sell artificial intelligence for the plant floor, and the technology genuinely works. What gets skipped in the sales conversation is the prerequisite. A model that recognizes an abnormal vibration signature needs months of recorded normal vibration to know what normal is. A plant with no instrumentation and no history cannot buy past that, which is why the honest first project is often sensors rather than software.
It is also not for every machine. Applying condition monitoring to equipment that is cheap, redundant and quick to replace costs more than the failures would.
In practice
Before approving any predictive system, ask what is currently measured on that machine and for how long it has been recorded. If the answer is nothing, the project starts with instrumentation and a waiting period.
Frequently asked
How much history does a predictive model need?
Enough to have seen the machine in its normal states, and ideally through at least one failure of the kind you want predicted. That is months rather than weeks, and it is the main reason these projects take longer than the demonstration suggests.
Which machines justify it?
The constraint, anything whose failure stops shipments, and anything whose failure is dangerous or slow to repair. For everything else a schedule or run-to-failure is cheaper than the monitoring.