DIY Predictive Maintenance
When is it worth installing vibration sensors?
You hear about predictive maintenance everywhere. At Industry 4.0 conferences, in system integrators’ brochures, in consultants’ slides. The message is always the same: stop repairing things when they break; start predicting when they will break.
All true. The problem is that most of the solutions on offer are designed for large-scale plants, with equally large budgets.
What about SMEs? What about those with a production line comprising 10 machines, a part-time maintenance technician and no R&D department? Can they implement predictive maintenance sensibly, without relying on enterprise platforms costing hundreds of thousands of euros?
The answer is yes — but with a few caveats. In this article, we explain when vibration sensors make sense, what they actually measure, and how to work out if it’s worth the effort for your situation.
What is predictive maintenance (and what it isn’t)
Predictive maintenance is based on a simple idea: machines send out warning signs before they break down. A worn bearing begins to vibrate abnormally weeks before it fails. A motor with alignment issues generates heat and irregular vibrations. A pump suffering from cavitation produces a recognisable vibrational signature.
Installing sensors means listening to these signals continuously and automatically, rather than waiting for the fault to become apparent — or for it to occur at the worst possible moment.
It is not, however:
- A magic solution that eliminates all faults
- Something that works without an initial learning phase
- Cost-effective for every piece of machinery
This last point is the one least often mentioned, but it is the most important.
Why vibrations?
Of all the physical parameters that can be measured on a rotating machine — temperature, current, pressure, noise — vibrations are the ones that reveal the most, in the shortest possible time.
A bearing that is deteriorating changes its vibration profile in a characteristic way that can be measured using accelerometers. An imbalance in the rotor generates vibration at the rotational frequency. Misalignment produces specific harmonics. Gear wear has a very precise frequency signature.
Temperature, by comparison, is a lagging indicator: when it rises abnormally, the damage is often already advanced. Vibrations predict failure, sometimes days in advance, sometimes weeks.
When is it really worth installing vibration sensors?
This is the key question, and it deserves an honest answer.
It makes sense if…
- The machine is critical to production. If it stops, everything (or almost everything) comes to a standstill. Even a single unplanned shutdown can cost more than an entire monitoring system.
- The fault has a history. If that machine has already broken down two or three times in the same place, you already have proof that the problem is recurring. The sensors become an investment, not an experiment.
- Repair times are long. Replacing a special bearing on an industrial gearbox can result in days of downtime. Anticipating a failure by as little as 72 hours allows you to plan and order parts in advance.
- The machinery is rotating and uses standard components. Motors, pumps, compressors, fans, gearboxes: these are the ideal candidates.
- You already have baseline data. Or you are willing to collect it. Predictive maintenance requires knowing how a machine vibrates when it is working properly, so you can recognise when something changes.
It doesn’t make sense (yet) if…
- The machine is economically replaceable. If a €200 motor breaks down, you replace it. Installing a €500 monitoring system isn’t a rational choice.
- Failures are unpredictable by nature. Breakages caused by impacts, sudden electrical events, human error: vibration sensors don’t predict these.
- You have no one to interpret the data. Vibration data, on its own, tells you nothing. You need thresholds, trends, frequency analysis. Without someone (or something) to process them, you’re just collecting numbers.
- The environment is too noisy. Excessive external vibrations (industrial floors, adjacent machinery, mechanical impacts) make it difficult to isolate the meaningful signal.
What exactly does a vibration sensor measure?
A MEMS accelerometer — the core component of most IoT vibration sensors — measures acceleration along one, two or three axes, expressed in g (gravitational acceleration) or m/s².
Metric | What does it measure? | A typical indicator of |
RMS | Average vibration amplitude over time | General condition of the machine |
FFT (frequency analysis) | Spectral profile of the signal | Specific type of fault |
Kurtosis | Impulse spikes in the signal | Localised defects in the bearings |
Temperature | Heat on the housing or bearing | Excessive friction, overload |
A practical framework to get started
You don’t need an enterprise IIoT platform to carry out predictive maintenance on a small scale. Here is a realistic architecture for an SME:
- Sensor level: a triaxial accelerometer with adequate sampling rate (at least 1–2 kHz for standard applications), rigidly mounted on the bearing housing. Communication via Bluetooth LE, LoRa or wired connection.
- Edge level: a microcontroller (STM32 or ESP32, depending on requirements) that acquires the signal, calculates basic parameters (RMS, simplified FFT) and transmits only the aggregated values, not the continuous raw signal.
- Cloud/local level: a server or cloud service that collects the data, displays it on a dashboard and sends alerts when thresholds are exceeded. This could be InfluxDB + Grafana, Node-RED, or a dedicated IoT platform.
- Human level: someone — even part-time, or the existing maintenance technician — who reads the alerts, interprets the trends and makes decisions. This role is essential, at least in the initial phase.
How much does it (really) cost?
A reasonable DIY system for a single machine can cost between €300 and €1,500 in hardware, plus the time required for installation and configuration.
Component | Approximate cost |
Vibration sensor (MEMS-based / industrial) | 50 – 400 € |
Microcontroller + enclosure | 30 – 100 € |
Connectivity and gateways | 50 – 200 € |
Software / dashboard (open source) | 0 – 100 €/month |
Setup, calibration, baseline | 1–2 working days |
The true cost, which is often underestimated, lies in the setup and calibration: establishing the baseline, setting thresholds, and testing the alert system. In a well-executed project, this is at least as important as the hardware.
A concrete example
Imagine a screw compressor used for 16 hours a day in a mechanical workshop. It has already broken down twice in three years, each time resulting in 3–4 days of downtime and around €2,000 in labour and spare parts. Fitting a vibration sensor to the front bearing, connected via LoRa to an existing gateway, costs around €600 in materials and one day’s labour. At the first fault alert, the maintenance engineer inspects the compressor, finds the bearing in the early stages of deterioration and replaces it in two hours during a scheduled shutdown. Production downtime: zero. Total cost of the intervention: €150. In this case, the ROI was achieved at the first anticipated event.
Conclusion: start with just one car
Do-it-yourself predictive maintenance doesn’t mean fitting sensors to everything. It means choosing the right machine, with the right criticality, and learning from it.
One sensor, one machine, six months of data. If it works — and it often does — you already have concrete evidence to take back into the company to expand the project.
At Omnetica, we help companies do exactly that: start with a real-world use case, choose the appropriate technology, and build the minimum necessary architecture. Without over-engineering, without unrealistic promises.
Is there a car that’s causing you concern? Let’s talk about it: we’ll give you our honest opinion on whether vibration sensors are the right solution.