Corrective, preventive, condition-based, predictive: on the ground these words get mixed up, and every vendor pushes its own. Here is the clear picture.
Maintenance falls into two families: corrective (you repair after the breakdown) and preventive (you act before). Preventive splits into scheduled (at a fixed interval) and condition-based (according to the machine's actual state). What is called "predictive maintenance" is an evolved form of condition-based maintenance: it forecasts degradation from signal analysis.
Mimorian is an industrial intelligence platform that models equipment, structures failure diagnostics and captures the know-how of maintenance teams through a multi-agent AI architecture. This article places each strategy, shows its limits, and explains why the right intervention threshold is built first from the facts and the knowledge of your teams.
Maintenance strategies, from corrective to predictive
The European standard EN 13306, which sets the vocabulary of maintenance, first distinguishes corrective maintenance from preventive maintenance [Source: EN 13306, 2018].
- Corrective maintenance: you act after the failure. The machine breaks, you repair. Simple, but the stoppage is unplanned.
- Scheduled preventive maintenance: you replace at a fixed interval, every six months or every 10,000 cycles, whether the part is worn or not. It is the right choice for safety components, where the risk forbids waiting.
- Condition-based preventive maintenance: you monitor a parameter (vibration, temperature, oil quality, number of cycles) and you act when a threshold is reached. You avoid replacing a part that is still good.
- Predictive maintenance: an evolved form of condition-based maintenance, which forecasts the moment of degradation from the repeated analysis of parameters [Source: EN 13306, 2018].
| Strategy | Trigger | When to choose it |
|---|---|---|
| Corrective | The breakdown | Low-criticality equipment, quick repair |
| Scheduled preventive | The calendar | Safety, parts with a known service life |
| Condition-based preventive | A threshold on the actual state | Equipment costly to stop |
| Predictive | A forecast of degradation | Instrumented machines, regular cycles |
Keep the hierarchy in mind: these strategies nest. Predictive is a form of condition-based maintenance, and condition-based is a form of preventive.
Predictive maintenance: what it actually does
Behind the word lies a precise mechanism: sensors measure a signal (vibration, temperature, current, oil analysis), a model learns the machine's normal signature, and it raises an alert when the signal drifts away from it.
The music analogy helps. You know the rhythm and the key of a piece. A wrong note is heard at once, and you guess that the string is about to snap. Predictive maintenance works the same way: it spots the departure from a learned normal.
Predictive maintenance does not predict a certain failure date. It estimates a window, a remaining useful life (RUL), with a margin of uncertainty. Marketing sells certainty, engineering sells a probability with a date attached: that confusion is what loaded the word "predictive" with hollow promises.
The method works, with conditions. Companies mature in predictive maintenance gain on average 9% in machine availability [Source: PwC/Mainnovation, 2018]. But it demands a fleet of sensors, a failure history rich enough to train the model, often a data analyst, and it mainly targets rotating machines with regular cycles, where the signature is stable [Source: ISO 17359, 2018]. On the first failures of a new piece of equipment, you sometimes have to run to failure once to learn the signature that precedes it.
Why predictive maintenance often boils down to condition-based
Here is the point the brochures leave out. Once the signal is analysed, prediction sets a threshold: above a given vibration level, you intervene. Setting a threshold on a parameter is the very definition of condition-based maintenance. The standard states it plainly: predictive maintenance is condition-based maintenance carried out following a forecast [Source: EN 13306, 2018].
So the "predictive" label sold as a breakthrough is, in most cases, condition-based maintenance with a finer trigger. The useful question lies elsewhere: where does the threshold come from, and does it cover what actually breaks your machines?
The blind spot of prediction: people and process
A model that reads sensors sees one thing: the degradation of a component. It does not see the rest, which weighs heavily in real stoppages:
- a setting or configuration error;
- a wrong part fitted at the previous intervention;
- operating the machine outside its limits;
- an instruction misunderstood on a particular piece of equipment.
In these cases, the component signal stays good, the sensor triggers nothing. Yet the line stops, because the problem lives in the process and in the human action. Human error accounts for 23% of unplanned downtime [Source: Vanson Bourne/ServiceMax, 2017]. Sensor-based prediction, on its own, stays blind to that share.
The Mimorian way: the right threshold, from facts and knowledge
Mimorian starts from the other end. Instead of instrumenting the machine, the platform uses what the plant already has: the failure history, the intervention reports, the reasoning of the technicians. It links symptoms to causes and remedies, and spots the components that fail again and again.
From this reading of the facts comes information the sensor does not give. When a part has been changed three times in six months, reliability is the issue. There you hold a real threshold: instead of waiting for the breakdown every six months, you plan the replacement at four months. The threshold comes from the machine's real history.
This path delivers the outcome expected from "predictive", anticipating before the breakdown, without the fleet of sensors or the data analyst, and starting from the existing data. It also covers the human and process blind spot, because it works on the real causes. Even a seasoned expert benefits: Mimorian generates intervention reports by voice dictation and makes schematics available in an instant, with no extra data entry.
Do you need sensors? The right combination
Prediction and the fact-based approach complement each other. The right architecture is simple:
- Foundation: guided diagnostics and knowledge capture across the whole fleet, including machines without sensors.
- Sensors on ultra-critical equipment, the ones whose downtime costs the most.
- Prediction backed by knowledge: when a sensor detects an anomaly, the AI cross-references it with the history. "Vibration anomaly. Based on past interventions, bearing or valve likely. Guided diagnostics recommended."
- Feedback loop: every intervention enriches the base, and the next diagnostic goes faster.
An example. On a hydraulic unit, scheduled maintenance runs every six months. Between two visits, a pump degrades and breaks in the middle of production. Between the part, the emergency intervention and the line stoppage, the bill quickly reaches several tens of thousands of euros. By cross-referencing the knowledge about this pump (known failure modes, history) with the monitoring of its condition, the AI spots the right moment to act, and gives the team the means to justify an early replacement, figures in hand. This intervention based on actual condition, tuned to the right moment, is the principle of condition-based maintenance. The same budget protects production better.
Frequently asked questions
What is the difference between preventive and predictive maintenance?
Preventive covers everything that acts before the breakdown. Predictive is a sub-category of it: condition-based maintenance that forecasts degradation from signal analysis [Source: EN 13306, 2018].
Is predictive maintenance the same as condition-based maintenance?
Very close. Predictive is condition-based maintenance with an added forecast. In both cases, you act when a threshold on the machine's actual state is crossed.
Do you need sensors to anticipate a breakdown?
Not always. Analysing the failure history and the teams' knowledge already yields useful intervention thresholds, across the whole fleet. Sensors add a fine layer on the most critical equipment.
Does Mimorian do predictive maintenance?
Mimorian reaches the outcome that prediction aims for, anticipating before the breakdown, starting from facts and knowledge rather than a fleet of sensors. On ultra-critical equipment, it combines with a sensor-based predictive layer.
Conclusion
Three points to remember:
- Maintenance strategies nest: predictive is condition-based, condition-based is preventive [Source: EN 13306, 2018].
- Sensor-based prediction alone stays blind to human error, configuration and process, where a large share of stoppages happens.
- The most reliable intervention threshold is built first on the facts and the knowledge of your teams, then refined with sensors on critical machines.
Next step: see how guided diagnostics structure the reasoning and capture the knowledge of your teams.
Going further: our guide to AI-guided diagnostics in maintenance, the guide to trustworthy AI for maintenance, the right moment for condition-based maintenance and, for the sensor-free version, the functional digital twin versus predictive maintenance.