An elevator motor begins to vibrate in a measurably different way two weeks before it fails. A compressor draws more current than usual three days before the bearing seizes up. Predictive maintenance uses such early warning signs to schedule maintenance based on a machine’s actual condition rather than on fixed calendar intervals.
Predictive maintenance cuts unplanned downtime by 30 to 50 percent and maintenance costs by 40 to 60 percent. The effort is worthwhile if the costs resulting from downtime amount to approximately 20,000 euros per incident or 2,000 euros per hour of downtime. A model requires 12 to 18 months of sensor data and documented outages before it can make reliable predictions.
Why This Topic Is on Every Maintenance Agenda
is the average cost of one hour of unplanned downtime. 67 percent of German industrial companies experience such downtime at least once a month.
ABB Analysis via b-und-i.de, 2024
A study by Fluke in collaboration with Censuswide estimates the annual losses incurred by German manufacturers due to downtime at approximately 44 billion euros (Fluke and Censuswide via industr.com, November 2025). Sensors measure vibration, temperature, current, or sound; a model compares the values with normal operating conditions and reports a deviation before the machine comes to a stop. The sources are listed in the Report from b-und-i.de.
What Predictive Maintenance Does
| metric | Change |
|---|---|
| Unplanned Downtime | 30 to 50 percent less |
| Maintenance Costs | 40 to 60 percent less |
| Overall equipment effectiveness (OEE) | 8 to 15 points higher |
These ranges can be explained by the initial state (KI-Syndikat, 2023). A company that has relied solely on reactive maintenance so far stands to gain more than one with an established preventive maintenance program.
The key lies in the advance warning time. If you know that a bearing will fail in three weeks, you can schedule its replacement during a maintenance window that was already planned. This saves you from having to make emergency repairs, pays off by avoiding rush-order surcharges for replacement parts, and often prevents consequential damage to adjacent components. Maintenance technicians often report that the number of nighttime emergency calls drops significantly as soon as the most critical units are being monitored. Ultimately, it’s the predictability that’s the real benefit—not the sensor technology itself.
When Predictive Maintenance Isn't Worth It
Predictive maintenance is not a standard tool for every machine. According to an analysis by Evarlink, the investment is only worthwhile when downtime costs reach approximately 20,000 euros per incident or 2,000 euros per hour of downtime (Evarlink, 2026). Below this threshold, the costs of sensors, modeling, and ongoing operation often outweigh the benefits. A simple ventilation system in an outbuilding does not require a predictive model.
There are technical limitations as well. A model requires 12 to 18 months of sensor data and documented failures to make reliable predictions (Industrieanzeiger, 2026). Machines that rarely fail, or whose faults do not manifest as vibrations, temperature changes, or current draw, do not provide a pattern that a model can learn from. Software errors, operator errors, or fluctuations in raw material quality fall outside this framework.
What a Pilot Project Needs
A pilot project involving a specific machine type typically costs between 80,000 and 150,000 euros in the first year, and the payback period is usually less than 18 months (KI-Syndikat and Groenewold, 2026). This includes sensors, connectivity, modeling, and support during the first few months while the system is still learning. Anyone planning to take this step should first determine which piece of equipment poses the greatest risk to production and whether there is sufficient historical data on past failures. If this data is lacking, the project will realistically begin with the data collection phase, not with the model.
As we mentioned in the post about the Implementation of Equipment Monitoring , the state of the data often decides a project's success more than the software chosen. That is exactly where our Plant Monitoring because it collects the sensor data that a predictive model needs in the first place.
What a realistic entry point looks like
- Taking Stock. Which machine has broken down most frequently over the past two years, which breakdowns were the most expensive, and are there even any records of this? In many companies, this data exists only in people’s heads or on paper, scattered across multiple shifts.
- Sensor Selection. Tailored to the machine’s expected failure patterns, not to a standard catalog. A gearmotor typically reveals problems through vibration, a heater through its temperature curve, and a variable-frequency drive through its current consumption. If you install every available sensor indiscriminately, you’ll collect a lot of data but gain little insight.
- Operational phase. The model learns, and false alarms are reduced. Experience shows that this phase lasts several months and requires someone at the facility to review the alerts and compare them with actual conditions. Without this feedback, the system remains ineffective, no matter how good the sensors are.
Is it worth it for your most critical system?Tell us the machine type and the estimated cost of downtime per hour. We'll let you know whether a specific model is needed or if a simpler measurement will suffice.
Frequently Asked Questions
How much does it cost to get started with predictive maintenance?
A pilot project covering one machine type runs between 80,000 and 150,000 euros in the first year, depending on the number of sensors and the complexity of the integration. Many plants recoup this investment within 18 months (KI-Syndikat and Groenewold, 2026).
What sensors are needed for predictive maintenance?
The most commonly used sensors are vibration sensors for rotating components, temperature sensors, current sensors, and ultrasonic microphones. The appropriate combination depends on the typical failure patterns of the machine in question. A gearbox exhibits different symptoms than a pump.
Does predictive maintenance also work with older systems?
Yes, retrofit sensors can be used to upgrade older machines that have no sensors of their own. The prerequisite is that impending failures announce themselves at all in measurable quantities such as vibration or temperature. For purely electronic or software-related faults the approach fails, no matter how many sensors are installed. How a retrofit works is described in the article on Retrofit Instead of Buying New.
The Next Step
torck develops its plant monitoring itself, with its own teams in Maxhütte-Haidhof, Vienna and Rabat, and knows from its own projects in industry and retail which sensor data a predictive model actually needs. Whether predictive maintenance pays off for your plants is best settled in an initial consultation. To arrange one, get in touch via our page on Plant Monitoring on.
German 
English 
French