Predictive maintenance: where it works — and where it is oversold
Condition-based maintenance, honestly placed.
Predictive maintenance uses condition data to forecast when a component will fail and schedules the intervention before it does. The concept is sound and, in certain applications, highly effective. It is also one of the most oversold terms in industry — because the preconditions rarely come with it.
The four maintenance strategies
Reactive means: repair once it breaks. Preventive means: replace on a fixed interval regardless of condition. Condition-based means: measure and intervene once a threshold is reached. Predictive goes one step further and forecasts from the trend when the failure will occur — which allows the intervention to be planned economically rather than merely in time.
The economic gain rarely lies in the avoided component damage. It lies in the avoided unplanned downtime and in bundling: if you know three components are due within the next eight weeks, you turn that into one appointment instead of three.
What it requires
Three things have to come together. First, sensors that actually capture the relevant condition — vibration, temperature, pressure, current draw. Second, enough history of real failures for a model to learn how a failure announces itself. Third, a process that can act on the prediction: technician, part and slot have to be available, otherwise the forecast is just earlier bad news.
The second point is the most common stumbling block. Well-maintained plants rarely fail — so exactly the events you would need to learn from are missing. That is not an argument against the method, but it is a reason to check the data situation honestly beforehand.
Where it pays off
Where unplanned downtime costs a multiple of the repair: continuous production, energy, process industry, rail, critical infrastructure. There the investment carries even when the forecast only catches a share of the failures.
It makes less sense for cheap components with low failure consequence, for very heterogeneous fleets without sensors, and anywhere the real problem is not prediction but diagnosis in the acute case. An operation that does not know which part is faulty during a live fault gains more from better remote diagnosis than from forecasting models.
Table of contents
Predictive maintenance forecasts failure timing from condition data and schedules intervention beforehand. It works where sensors, failure history and high downtime cost meet — and fails where those foundations are absent.
FAQ
What is predictive maintenance?
A maintenance strategy that forecasts the likely failure time of a component from condition data and schedules the intervention beforehand — instead of acting on a fixed interval or only after the failure.
What data does predictive maintenance need?
Condition data from sensors such as vibration, temperature, pressure or current draw, a reliable history of actual failures, plus context data on operating hours, load and maintenance events.
When is predictive maintenance not worth it?
For cheap components with low failure consequence, where sensors and failure history are missing — and where the real problem is diagnosis during an acute fault rather than prediction.
We don't just write about it. We build it.
Most companies don't fail on ideas — they fail on implementation. We partner with teams to turn methods like this into concrete outcomes: clearer choices, faster delivery, measurable impact.
