Maintenance
Condition-based maintenance
How a job driven by a degradation measurement works, and how to choose the variable, the threshold and the tolerance to spikes.
In condition-based maintenance the job is not decided by a date or by a counter, but by a measurement: you intervene when a quantity shows the component is genuinely getting worse. It is the most accessible form of predictive maintenance, because it needs no complex models, only a signal that comes before the failure.
The three ingredients
The variable. A quantity measured on the machine that correlates with the degradation of the component: the vibration of a bearing, the temperature of a motor, the differential pressure across a filter, a physical property of the working fluid. Correlation is the point: a measurement that moves for a thousand other reasons is of no use, even when it is available.
The threshold. The value beyond which the job becomes necessary. It can be an upper limit (vibration going up) or a lower one (a fluid losing viscosity).
The tolerance to spikes. How long the condition has to stay true before it counts as real. Without it, a transient of a few seconds creates a job nobody would have planned, and after three false alarms the plan gets switched off.
When the condition occurs and persists, the job enters the maintenance plan and the people responsible are notified. It is not a machine lockout: it is a job planned a few days ahead instead of a sudden stop.
Examples of logic
Three typical cases, given as the structure of the reasoning rather than as numbers, which always depend on the component and on the manufacturer:
- bearing: effective vibration goes above the level considered acceptable for that class of machine and stays there for tens of minutes, not for an instant. The job is an alignment check or a planned replacement;
- hydraulic fluid: a property of the fluid falls below the limit stated by the manufacturer, and stays there. The job is the change, brought forward or pushed back with respect to the calendar due date depending on what the measurement says;
- drive: motor temperature stays above its normal working value even at constant load. The job is a check of tension and wear, before the belt gives way in production.
In all three cases the value of the threshold does not come from the application: it comes from the manufacturer specifications, from the reference standards for that type of machine, or from the history of the plant itself.
How to find the right threshold
This is the hard part, and the reason it is worth measuring before raising alarms.
The soundest route is to watch the quantity for a few weeks of normal operation, work out where its usual band lies and how high its legitimate spikes go, and set the threshold above that band. A threshold dropped in from outside onto a machine that has never been measured almost always produces one of two outcomes: too many alarms, or none at all.
The opposite of common belief also holds: an over-cautious threshold is not the safe choice. It creates unnecessary jobs, burns credibility and ends up being ignored, which is the worst place to be when the measurement really does have something to say.
The prerequisite: the data has to be there
A condition can only be evaluated if the quantity arrives continuously. If the sensor stops transmitting, the condition does not become false: it becomes impossible to evaluate, and the component is left uncovered without anyone noticing.
This is why condition-based maintenance is always worth pairing with a periodic safety due date, as described in maintenance types: the measurement is there to bring the job forward when it is available, not to justify its absence when it is missing.