Predictive maintenance is often presented as a high-tech future of AI models and thousands of sensors. For small and mid-sized factories, the reality is more practical — and more useful. Predictive maintenance simply means maintaining equipment based on its measured condition rather than on fixed intervals or after it breaks. Done selectively, on the right assets, it prevents costly failures and avoids unnecessary maintenance.
Three maintenance strategies
- Run to failure: fix it when it breaks. Acceptable for cheap, non-critical items with spares available.
- Preventive maintenance: service at fixed intervals of time, cycles or run hours. Good for wear that correlates with usage.
- Predictive (condition-based) maintenance: monitor condition and act when it deteriorates. Best for critical assets whose failures are random in time but show warning signs.
A sensible maintenance plan uses all three, chosen per asset.
Which assets justify predictive maintenance?
Use a simple criticality review: what happens if this asset fails? Consider production impact, safety, quality, repair cost and lead time for spares. Assets with high impact and detectable warning signs are candidates. Typical examples in SMEs:
- Main air compressors and vacuum pumps.
- Large motors, gearboxes and fans on bottleneck lines.
- Hydraulic power packs on presses and moulding machines.
- Spindles on critical CNC machines.
- Chillers, cooling towers, boilers and transformers.
See criticality and spare parts strategy.
Condition monitoring techniques
Vibration
Rising vibration on rotating equipment indicates imbalance, misalignment, looseness or bearing wear. Simple overall vibration levels trended over time catch many problems; spectrum analysis identifies the specific fault.
Temperature
Bearings, motors, electrical panels and hydraulic oil heat up as problems develop. Infrared thermometers, thermal cameras and fixed sensors all have a place.
Motor current
Changes in current draw can indicate mechanical load problems, blocked filters or electrical faults. Current sensors are often easy to install in panels.
Oil analysis
For gearboxes and hydraulics, oil samples reveal wear metals, contamination and oil degradation — often months before failure.
Run hours and cycles
Not strictly predictive, but triggering maintenance on actual usage rather than calendar time is a big step forward for machines with variable utilisation.
Starting small
- Pick five to ten critical assets.
- Choose one or two parameters per asset that indicate its most likely failures.
- Start with periodic readings — weekly vibration and temperature with handheld instruments, recorded on a mobile app.
- Set alert thresholds based on manufacturer guidance and your baseline readings.
- Create work orders automatically when readings cross thresholds.
- Add fixed sensors where readings are frequently needed or access is difficult.
- Review results — failures caught early, failures missed — and adjust.
An example: compressors and a hydraulic press
A plastics processor depended on two screw compressors and a large hydraulic press. A compressor failure in the previous year had stopped half the plant for two days. The maintenance team began weekly vibration and temperature readings on the compressors and monthly oil samples on the press hydraulics, recording them on a tablet with simple trend charts. Within a few months, rising vibration on one compressor's motor bearing triggered a planned bearing replacement during a weekend — avoiding a breakdown. Oil analysis on the press showed contamination from a failing seal, replaced before it damaged the pump. Later, the company added fixed temperature and current sensors to the compressors so alerts arrived automatically.
Integrating with your maintenance system
Condition data is only useful if it leads to action. Readings, trends and alerts should live in the same system as work orders, PM schedules and spare parts. When a threshold is crossed, a work order should be created with the reading attached, and the technician's findings should be recorded against the asset — building the history needed to refine thresholds. See CMMS software.
Realistic expectations
Predictive maintenance doesn't eliminate all failures, and sensors themselves need maintenance. Some failure modes give little warning. The goal is to catch the most damaging, detectable failures early and to stop doing unnecessary preventive work on assets in good condition. Third-party sensor, gateway and platform costs are paid by the client, so investment should follow criticality.
Common mistakes
- Instrumenting every machine before proving value on a few.
- Collecting data without thresholds or anyone responsible for acting on alerts.
- Dropping preventive maintenance entirely in favour of sensors.
- Ignoring false alarms until people stop trusting the system.
How software helps
Maintenance software brings readings — manual or from sensors where available — together with assets, thresholds, alerts, work orders and spare parts. Trends and failure history show which assets and techniques deliver results. See IoT manufacturing platform.
Talk to us about condition-based maintenance for your critical assets.
Building the business case
Estimate the cost of one failure of each candidate asset: lost production hours multiplied by contribution per hour, repair cost, expedited spares, overtime and any quality or customer impact. Compare this with the cost of monitoring — instruments or sensors, installation, platform subscriptions where used and the time to review data. For critical assets, a single avoided failure often pays for monitoring several times over. For low-criticality assets, it rarely does, and preventive maintenance or run-to-failure remains the better choice.
Getting started
List your ten most critical assets, decide one or two condition parameters for each and start weekly readings with handheld instruments. Record them in your maintenance system with thresholds, and review trends monthly. Add fixed sensors only where the readings prove valuable.
Frequently asked questions
What's the difference between preventive and predictive maintenance?
Preventive maintenance happens at fixed intervals of time or usage. Predictive maintenance acts when measured condition — vibration, temperature, current, oil quality — shows that a failure is developing.
Is predictive maintenance only for large plants?
No, but it should be targeted. SMEs get the best return on a small number of critical assets where failures are costly or dangerous.
Do we need AI for predictive maintenance?
Not to start. Simple thresholds and trends on vibration, temperature or current catch many developing failures. Advanced analytics can come later if data justifies it.
What sensors are typically used?
Vibration sensors on rotating equipment, temperature sensors on bearings and motors, current sensors on motors, and pressure or flow sensors on hydraulics and utilities. Sensor and platform costs are borne by the client.
Can we start without sensors?
Yes. Periodic readings with handheld instruments, recorded on a mobile app with trends and alerts, are a valid first step.
